An audiophile friend gave me a book for Christmas, written by Robert Harley, "The Complete Guide to High-End Audio". The quote below, which explains what I hear quite adeptly, is from that book, pg 231, from the section titled, :"16 Bits, 20Bits, 24 Bits":
"...the benefits of the increase in word length from 16 to 18, 20, and even 24 bits are not in dispute. As I mentioned earlier, word length is the number of bits used to encode the audio signal's amplitude at each sample. Assigning a number--called a word--to represent the audio signal's amplitude is called quantization. The word length determines the system's resolution, dynamic range, distortion, and signal/noise ratio. We also call the word length resolution. . .
....the greater the number of bits in each digital word, the more precisely the analog signal's amplitude is encoded. . . The longer the quantization word, the more steps, and thus the finer the resolution.
. . . .the resolution of a digital audio signal isn't defined by the maximum number of bits available, but the by the number of bits being used at any given moment.
The advent of 20- and 24-bit digital audio not only expands the dynamic range, but also increases the resolution of low-level detail. This low-level detail can be fine nuances of an instrument's timbre, which enhance the sense of realism. It can also be subtle spatial cues, such as discrete acoustic reflections and reverberation decay, which the ear interprets as a more convincing reprodution of the original recording venue.
Longer word lengths also contribute to better sound because the postproduction (mixing, equalization, signal processing) common in the recording or mastering studio can be performed with much greater mathematical precision. Moreover, any noise added by these processes is spread out over a wider bandwidth, which makes it less audible.
The combination of higher sampling rate and longer word length results in greatly improved sound quality. (A digital system's resolution can also be increased by adding to the signal a small amount of noise, called dither...)"
He then proceeds to outline the limitations of 20+bit processing, often calling the additional bits, simply marketing bits, fi they're not done right, and that for most systems any information over 20 bits is "rarely capable of delivering real audio information"
So, I needless to say, it appears simple, but it quickly becomes a complex issue.