Three main reasons that I can think of, some of which has been covered here:
(1) What live2496 said. I've owned at least one "audiophile quality" DAC (praised to the skies by Stereophile magazine, for example) that clipped quite audibly on steady near-full-scale tones in the midrange, e.g. 600 Hz. I'm not sure that it's just "older" DACs, either.
It's also not just DACs. The "aux" inputs of some hifi equipment can't take the full peak output voltage of a DAC or CD or DAT player. On the other hand, that's a lousy reason to under-record--go buy a pair of 6 dB attenuators from Harrison Labs (
www.hlabs.com) and put them at the inputs to your preamp or receiver if you have this problem.
(2) The official Red Book specification (Compact Disc Digital Audio) originally said (and may still say, but I haven't read it since the early '80s) that the lowest-order four bits in a sample may not have the same value as the higher-order twelve bits if the higher-order twelve bits are all the same.
In other words, the range doesn't quite go all the way from -32768 to +32767 (in integer terms); the 15 lowest and 15 highest values are technically "off limits." The dynamic range that is given up when this rule is followed is less than 1/1000 dB, however, and I don't believe that the rule was ever taken very seriously in practice. I've never heard of a CD pressing plant rejecting a tape master for this reason, for example.
(3) Theoretically at least, DACs need somewhat greater headroom than some people might realize--their peak output voltage when playing back a full scale "steady tone" (sinusoid) isn't
quite the absolute maximum that they would need to put out with certain peculiar sample value sequences that can be synthesized.
The overload in such a case could be as great as 4 - 6 dB. But it would last only a fraction of a millisecond if it ever actually occurred--and given the mathematically deterministic nature of the sample sequences required to trigger this effect, acoustically recorded music would essentially never contain such a signal except by an almost unimaginably freaky accident. Still, some poorly implemented audio circuits "ring" (oscillate) when they're overloaded even briefly, such that the secondary effects might be worse than the actual clipping.
--best regards