I'm not too sure either what it is doing with 16/44.1 signals, but had one guy post this re the subject:
I only have 44.1/16 music and now upsample it all to 176.4 (rather than 192 as 176.4 is a whole number multiplier of 44.1) using Secret Rabbit Code Level 0 (the highest quality) all performed in software within mpdPup to a WaveIO. It is excellent and gives a major boost to SQ without IMHO any unpleasant artifacts.
another guy's post:
I think that might have to wait. That I haven't switched off your DAC after a few hours listening and fired up my TDA1541A comfort blanket, says it all really. Don't quote me on this, and of course I am allowed to change my mind if I ever do implement a 1704 daughter-board, but my gut feeling is that 1704 will sound more "digital" than what you have now. There is a naturalness, a kind of organic nature, which reminds me more of old-school Burr Brown and Philips chips, more than anything else I have heard which has a sigma-delta badge, or even 1704 for that matter.
I will post some more later. But I had a most interesting time last night with mpd upsampling 44.1/16 CD rips to 176.4. (I don't have a great deal of >44.1k material, so was just playing with software upsampling redbook to 88.2 and 176.4 on the laptop.) Since then I have installed Triode's EDO plugin and kernel on my Logitech Touch (for USB DAC support and > 96k sample rate support) and currently have the DAC hooked up to that. The idea being that I could after editing a config file on the server, use sox or ssrc, to up/over sample on-the-fly. Only problem with that is that I can't go above 96k at the moment due to the Touch using wireless network. Doesn't have the bandwidth to support 176k sample rates. I'll hook up via ethernet later. Reading the John Swenson post, (linked above), I think he is onto something. If you are going to OS and DF, you can get better results doing it in software, not the filters built into the chips.