Originally posted by legion
It's a bit of a long explaination...
The crux of the problem is a phenomenon called jitter which refers to small timing errors that can be introduced as the digital bitstream is read off the disc and transported through various processing stages on both the source component and the receiving component. The word clock that cranks out bits in quick succession as tiny electrical pulses can drift by several tens or hundreds of picoseconds between clock periods. The explanation is more complicated than this, but the result is that jitter produces undesirable audible effects. Jitter cannot be completely eliminated, but it can be significantly reduced. Apparently, coaxial connections offer more flexibility in controlling jitter.
In addition to coax and TOSLink, there are some proprietary interconnect formats such as Laser Linque (aka Single Mode) that is a more advanced version of TOSLink (i.e. less jitter). I think it's the comparatively larger diameter of TOSLink that reduces the ability to control jitter. Laser Linque specifies a significantly smaller diameter along with other enhancements.
Robert Harley, formerly of Stereophile magazine, has written a detailed explanation here (I haven't read it all..yet):
http://www.stereophile.com/reference/368/index.html
The first two paragraphs:
"Not that long ago, digital audio was considered perfect if all the bits could be stored and retrieved without data errors. If the data coming off the disc were the same as what went on the disc, how could there be a sound-quality difference with the same digital/analog converter? This "bits is bits" mentality scoffs at sonic differences between CD transports, digital interfaces, and CD tweaks. Because none of these products or devices affects the pattern of ones and zeros recovered from the disc, any differences must be purely in the listener's imagination. After all, they argued, a copy of a computer program runs just as well as the original.
As our knowledge of digital audio has become more sophisticated, however, we've learned that the timing of those ones and zeros is of utmost importance. It isn't enough to get the bits right; those bits have to be converted back into music with the same timing reference as when the music was first digitized. It turns out that timing errors in the picosecond (ps) rangethe time it takes light to travel inchescan audibly degrade digitally reproduced music. These timing errorscalled jitterare only now beginning to be understood (footnote 1)."