What's better, one conductor or two conductors for an RCA interconnect?

When shield is used as return any electrical noise current induced in the shield will create voltage difference along it  - between source/output and destination/input.  Input will "see" the difference between output signal wire and output GND, plus noise voltage between both ends of the shield.

The best is to use two wires (signal and return) with shield connected only at one side.  

 

My understanding is it is better for electrostatic noise rejection but there may be circumstances where shielded twisted pair is better for magnetic.

Twisted pair exposes both wires evenly to electric and magnetic fields causing identical induced noise currents that cancel. It is extremely efficient, as long as the pitch of the twist is much smaller than wavelength of the offending electrical noise. Shield works great at higher frequencies where, in spite of being non-magnetic, shields by means of skin effect (noise currents flow on the outside - shield). Shielded twisted pair is the best combination.

As for grounding the shield - I would always ground it at the source and never at the receiver end only. The reason for this is that induced common mode electrical noise (identical in both wires) normally would cancel, but since both wires have impedance and slightly different distributed capacitance to shield, grounding it at the receiver creates two lowpass filters at two different frequencies converting common mode to normal mode signal. We’re talking of few picofarad difference, but at very high frequencies it can makes a difference. Also, intuitively I wouldn’t drain electrical noise from the shield into more sensitive side (input) device.

I suspect that reason for oscilloscope to have plain coax is capacitance. Twisting wires reduce inductance but increases capacitance - important with scope’s high input impedance and very high measured frequencies. Unfortunately this coax arrangement creates errors. When you short probe and touch tested circuit with shorted leads it will show small amount of noise (in spite of being shorted). It is because current flows from the circuit thru the shield (finding return to GND) causing voltage drop, that shows on the input as signal - exactly what we try to avoid in interconnects by using two wires inside of the shield.

Because twisted pairs provide cancellation to all external fields (magnetic or electric) it is used everywhere - in all network cables, in all audio cables etc. It would be stupid not to. Typical twisted pair provide about 40dB rejection up to about 100kHz and still 20dB rejection at 700kHz. Above that shield becomes very effective by means of skin effect. Skin depth changes with frequency squared, being roughly 2mm at 1kHz (Cu or Al). For 100kHz it will be 0.2mm. There is some info on twisted pair effectiveness here:

In single ended connection we are passing signal and reference point (analog ground). There is a loop from the output to input and back by return - analog ground. Induced electrical noise currents in both wires flow in the same direction and cancel. This would work perfectly if there is no other path for return, like when chassis on one side is not earth grounded (II class) or when it is grounded but analog ground is floating on either side. Analog ground is often connected to chassis ground with the large resistor. That would diminish effectiveness of twisted pair a little but still, twisting would help and perhaps that’s why many manufacturers twist single ended wires inside of the shield.
What balanced connection bring is complete symmetry and independence from this additional return path, but also, when properly done (signals not referenced to ground) removes effect of wires to shield capacitance (allowing for longer connections).

Another example of single ended output, that behaves like balanced, is amplifier output, since speaker is floating.

@deludedaudiophile So, you are not EE.  Well, I am - designing low level electronics for 40 years.  Sorry I cannot explain it to you better.  We agree to disagree.

Speaker by definition is balanced.  Noise currents induced in both wires cancel at the speakers, since there is no other return path.  If you question that, then perhaps you should read what Benchmark Media posted on their website in regarding to headphones being always balanced:
 

Headphone transducers respond to the voltage difference between the two wires that feed them. They have perfect rejection of common-mode interference because there is no path to ground or to any other conductor. In other words, there is no path for ground loops.

Headphone transducers are electrically isolated from everything other than the two wires that feed them. It doesn't matter if both conductors are driven differentially or if only one conductor is driven. The headphone transducer will reject common-mode noise.

The same happens when analog ground is floating on one side - no additional return path, hence inherently balanced connection with single ended output.  Analog ground is likely to be connected to chassis ground to reduce noise coupled from the chassis ground to circuitry, but it is often done with resistor of 100k or more.  It is because connecting analog ground to chassis on both ends directly or thru large capacitor creates perfect ground loop (earth ground - chassis - analog ground - interconnect - analog ground - chassis - earth ground).  That is why Benchmark amp doesn't even have single ended input.   This additional return path, possibly thru (two) 100k resistors creates some unbalance, but doesn't completely null usefulness of twisted pair in SE connection and that is why manufacturers use it.  For speaker wires it is no-brainer since it reduces noise, reduces inductance (important) and increases slightly capacitance (not-important).

As we continue about our qualifications, I assume that you've never designed any electronics?  As for me - I don't have PhD (only MSEE), meaning I'm still able to learn  :)    One more thing (Columbo) - it is "kHz", not KHZ.  I don't want to be unit police but I write it proper way automatically and it bothers me a little, when educated people don't pay any attention, sometimes using even erroneous terms like "watts rms"

As I clearly said, from the speaker standpoint, the speaker itself is balanced, however, if you are considering noise on the speaker cables induced back into the amplifier, it most definitely is not balanced because the impedance on either cable is not the same as the other. 

Impedance between what and what?.   Induced noise current flows in the loop  output - speaker wire - speaker terminal - speaker impedance - speaker terminal - speaker wire - output impedance - output.  In this loop noise currents will cancel, even if you earth ground it at one point (as long as it is one point only).  It is just simple loop with pickup (induced current) proportional to area between wires. 
How, on earth, noise currents IN THE LOOP can cancel for the speaker only and not for the amplifier.  Loop currents either cancel or not.  If you claim that currents in one leg/wire is different because impedance to ground is lower, then they wouldn't cancel for speakers as well.  How induced noise current in one wire can be different than current in another wire if it is loop with wires exposed evenly to external field, when loop impedance is same for both?   Grounding this loop in two places is different story because there will be additional return path.   Sorry I cannot explain it better.

Perhaps mentioning "KHZ" is pedantic, but it make me suspicious about your experience in electronics.  None of my fellow engineers would write it like that.  As for "deflecting" from technical discussion - it deflects less than bragging about PhD title.
(to be really pedantic:  it is 16k  not 16K)

If the current goes the other way, one side sees the impedance of the amplifier. The other (ground) sees the the impedance of the amplifier, but also sees a parasitic path through through that ground and out to "somewhere". Now you have a differential voltage caused by the common mode noise.

What do you mean current goes the other way?  It is the same current - not two different cases.  If it cancels at the speaker (no current thru speaker) it means net current in the loop is zero.  It will also be zero on the amplifier side (no current in the wires).   Impedance in the loop is the same for both wires - you cannot separate them - wires are in series.  Presence of the speaker doesn't change anything - we can short it.  It is the same wire where two induced currents flow in the opposite direction.  What might flow to ground is differential current that is zero. 

Draw two exactly same current sources of opposite polarities in series and close the loop with resistor.  Current in the loop, as well as voltage across resistor will be zero.  You can ground it at any point and it won't make a difference - it is floating circuit (no reference to anything).  Only connecting it at two points will unbalance currents (alternative path). 

As for the case of phono cartridge - as long as it is grounded on one side only it is still loop.  When you ground one end of the cartridge it won't make a difference as long as the other end is floating, for instance transformer.  It is the same case as with our headphones example - cartridge is inherently balanced and twisting wires will make sum of induced currents zero.  If it is zero at the cartridge it will be zero at the amp - no matter if and where you ground one of the wires.  Advantage of balanced input will be rejection of common mode noise picked by the cartridge.  Perfectly twisted wires will offer great common mode rejection in either case - balanced or single ended.

There are other reasons for driving headphones with balanced drive.  One is to get double voltage (often needed with 600ohm headphones), another is to avoid crosstalk between channels by using common return (GND).

@jumia   Sorry, for hijacking your thread - I won't write anymore.  To your question I can only say what I think is right.  I would use two wires with shield connected on the source side only.  Many SE cables have an arrow pointing direction from source to receiving end.  You can also unscrew shells to see and mark the end with shield soldered to case (GND).  

@atmasphere AFAIK we should write symbols of units named after person with capital unit letter V, kV, A, kA, kHz, kW etc. but spelled out units with lowercase : volts, amperes, ohms, watts.  Multiplier doesn't change that, so it will be kilovolts, kiloohms, kilowatts, kilohertz.  The only exception is when unit contains two words like degree Celsius.   

 

@jumia   At one RCA connector shield and negative/return wire are both soldered at the base of the connector while signal/positive wire is soldered to center pin.  On the other side only two wires are soldered while shield is left unconnected/floating.