The analog output circuitry of some playback equipment--perhaps more than you would expect, and including some expensive, highly-touted equipment--can overload briefly even when signal peaks are somewhat below 0 dBFS. By leaving just a small amount of headroom (a dB or two), you can reduce both the likelihood and the severity of any such distortion without sacrificing a quiet recording.
Part of this problem is due to the way digital anti-aliasing filters are often implemented (they don't necessarily have unity gain), and part of it is a very interesting phenomenon which can cause the signal within a sampling interval to exceed the peak amplitude of a 0 dBFS sinusoid even in a unity gain arrangement.
When you have a 90+ dB dynamic range to work with, there's simply no need to push that last dB or two.
--best regards