Glai 12-06-11:An ideal 10kHz square wave, having infinitely fast edges (rise times and fall times), consists of a summation of sine wave components at frequencies of 10kHz and all of its odd-numbered harmonics (out to infinity), each of them having different weightings that decrease as their frequency increases. The first odd numbered harmonic of 10kHz, aside from the 10kHz "fundamental frequency" itself, is 30kHz, which we essentially can't hear. I say "essentially" because some seemingly credible studies have reported that subjects are able to sense the presence of frequencies possibly as high as a few ten's of kHz above 20kHz, when those frequencies are present in combination with lower frequencies.
In reviewing the measurement sections of stereophile, many amps with -3db at 100khz demonstrate subtle rounding of the edges when reproducing 10khz square waves. I don't listen to square wave so I don't know what that translate into.
My basic point being that since risetimes and falltimes inversely correlate with bandwidth, our hearing mechanisms, not to mention the speakers, the source material, and the source component, will round off high frequency square waves to a considerably greater degree than most amplifiers will. With the possible (and likely very subtle) qualification suggested by the studies I mentioned, even if the 10kHz square wave were rounded off to a pure sine wave, we would not be able to hear the difference, assuming that the round-off effect is not accompanied by other side-effects.
On the other hand, a reasonable degree of bandwidth overkill in an amplifier will prevent that component's bandwidth limitation from contributing significantly to the overall bandwidth limitation of the system + our ears. As well as eliminating or minimizing other effects that could be manifested below 20kHz, some of which the others have noted above.
Again on the other hand, ultra-wide component bandwidth has potential issues of its own, as you correctly alluded to, including increased susceptibility to rfi, the possibility that some components in the chain may not be able to gracefully handle ultra high frequency components that are input to them by the preceding component(s), and the possibility that transmission line effects which occur at rf frequencies could cause signal reflections to enter feedback loops.
The bottom line, as I see it, is that in itself that spec does not have a great deal of usefulness. Although FWIW I would have to say that 100kHz strikes me as at or somewhat below the lower limit of my intuitive comfort zone, and all other things being equal I would give preference to a somewhat higher number.
Regards,
-- Al