For digital, you just have to pass the eye-pattern level for proper digital signal retention. The more open the "eye" the fewer error bits, the lower the jitter. This should be the full channel, too. Cable and connectors both. You don't need expensive cables to virtually remove bit errors. Example, Ethernet can transmit signal 328 feet at 10 MB/sec with virually zero dropped bits with 100-ohm 4PR24 23 AWG wire (250 Meg per wire pair times 4 ). Once you go digital, and with sufficient eye pattern margin, the "sound" is the AD converter, not the bits...which have no sound at all. I would look more at your AD converter than the cables, and such short cables seldom have issues with data rate Xmission. What's a few feet (with proper impedance matching) at best compared to 328 feet / 100 meters?
Shielding is only as important as the environment is poor. Mosy twisted pair digital cable has plenty of CMRR ratio to passively remove noise. But, a shield will attenuate the artifacts so that the remainder of the noise common mode rejection doesn't remove is 20-30dB smaller in magnitude. The shield doesn't improve the CMRR of the system, it just makes the noise the system has to deal with SMALLER. The reduction is the same "something" times X of the orignal signal, the original signal is just smaller.
Don't be fooled with shields, though. A shield concentrates the ground plane tightly around the cable which also magnifies a cables defects. Poor quality UTP cable that works, can be a disaster with an added shield. Once you BEND the cable, the issues keep getting worse, too. So shielding isn't to be taken lightly as an improvement unless you are seriously in the know about the cable under that shield. Most will have better performance with unshielded cables that are twisted pairs. Coaxial cables do better with shield (and have to have one to work, anyway) since the internal construction is easier to manage as it is a round dielectric and holds the shield at a constant distance from the internal unbalanced signal wire, and while the cable is bent, compared to balanced pair cables. Issues with noise are much less a problem with balanced cables that remove the noise through CMRR (RF) and the pair twists (low frequency magnetic interference), and good coaxial cables can put RF noise 100 dB below the signal but are more helpless against magnetic interference when used at audio frequencies below 1 MHz or so.
The typical wires to meet digital Ethernet eye patter and attenuation to cross talk margins (ACR) are sufficiently small to make "skin effect" not much of an issue. The 0 dB (where the signal is larger than the noise) ACR value extends past 500 MHz on good cables. Common electronics can see a signal as much as 6 dB below the 0 dB ACR frequency, too. So "finding" the digital stream isn't too tough, what's tough is the DA conversion at each end. All the advances in Ethernet technology trickle down into other digital media as it becomes affordable.
Kijanki et all are right on that the "system" length is important as you get shorter lengths as the reflected energy caused by impedance mismatch at SPDIF frequencies can be pretty large. On longer cables, reflected energy is attenuated out. Most all digital systems have a much harder time passing "short length" channels than long length channels for that reason. RL (Return Loss) is not so good with RCA connectors as they aren't 75-ohms. So digital audio seems to gets a deserved bad rap on qualty of the interface far in excess of the cable between the connectors.
It is also true that to have a "transmission line" in the true sense, the wavelength nees to be at least 1/4 length relative to the velocity of propogation of the cable. Shorter than that and you don't get reflections (remember the open and closed menometer physics experiments?).
Audioengr point number two is somewhat confused. Dielectric is called a di-electric because it apposes electric flow. This is true because it sits between two conductors and makes a capacitor, which apposes signal change or "flow". The capacitance is a fixed figure when you decide a cable's impedance which is 101670 / (capacitance x velocity of propagation of the dielectic). There is no "absorption" at all based on the capacitance, just the storage and release of energy. So if you want a faster cable at a given impedance, lower the capacitance by changing the dielectric.
101670/ (20.5 x 66) = 75 ohms
101670/ (17.3 x 78) = 75 ohms
Now, what does grab away the energy is the dissipation factor or loss tangent (same thing) of the material. The transverse electromagnetic wave energy transmitted between the two wire (view it as a wave between two plates), and traveling in the dielctric medium, is a reactive vector, and the "real" part verses the "imaginary" part create inefficiencies that causes lost energy. The imaginary (the loss "tangent") part is lost.
He's 100% correct that a vacuum is the best dielectric (or air as the poor man's vacuum) to lessen losses. Teflon does NOT need to be used, however. PP or PE is cheaper, and better than Teflon at normal room temperatures. Teflon is just expensive, and you can listen to your stereo in a 200F room if you like! The cable won't melt. Oh, Teflon is pretty, though.
Be wary of mixed conductors though. With such short distances the advantages are not significant. Yes, the higher frequencies attenuate more, so higher order digital frequencies should arrive less attenuated through silver...but at what length? And at what skin depth? I'd worry a whole lot more about the connectors and DA converter. Those pretty silver cables may look like you care more, but the hidden stuff no one sees matter more.
