After looking at the code for MPD volume control, it appears to be a very high quality algorithm, including the application of dither. http://git.musicpd.org/cgit/master/mpd. ... Volume.cxx
I’m thinking that MPD volume and digital hardware volume (USB or DAC chip based) should be similar in quality as long as hardware volume is using dither, or dither is applied at some point in the signal path through the chip. Dither is important for 16 bit samples but not so much for 24 bit samples since it affects only the LSB, and for 24 bit samples LSB = -144dB which is below the noise floor of the device.
Another observation is that chip based hardware volume implementations often do not specify much about their volume control processing for example, Tenor TE7022L and ESS Sabre 9012/9018.
http://www.gfec.com.tw/documents/10157/ ... PB-v21.pdf
http://myl8test.files.wordpress.com/201 ... t-here.pdf
Other chips for example, WM8742 provide a lot of detail including selectable dithering algorithms for volume.
http://www.wolfsonmicro.com/documents/u ... WM8742.pdf.