Earlier quoted context omitted.
If your YT/Winamp/whatever volume isn't at 100%, then what you're doing is: 1. generate audio signal 2. reduce volume of that signal, losing information because it's quantised 3. take that volume-reduced signal and boost it right back up again, but now with the lower bits destroyed You can make this effect as bad as you like, e.g. turn it down to 1% and then amplify by 100x... but why?
Because in order to get the best dynamic range while listening to diverse music, you need to detect the loudness of tracks and albums and adjust the gain of each track/album accordingly. In order to have room for the music player to adjust the volume higher for quieter tracks, you need to apply a "pregain" to lower the volume overall, while turning up your speakers to compensate. This solves the problem, but by doing…
I'm aware of ReplayGain and this processing is important for per-track overall gain, but what I'm getting at is lower level: instead of there being two lossy/rounded stages of dimming and amplification, you want to communicate to the OS a log2 "dimming factor", so that this can be subtracted from a later log2 amplification factor such that we ideally waste no processing time if the sum is zero, and otherwise don't suffer the twice-quantised signal degradation (at most one accurate scaling pass, instead of two arbitrarily precision-reducing ones). It's maybe a minor point / imperceptible as others have noted, but IMO this seems like the Correct (TM) approach.