it would suck to hear above 22k, damn, mine stops at like 15-16k (yeah, i know, thats really bad, i blast things too loud, i know, i'm going to be deaf by like age 60, i know, but then i get that awesome bone implant thing that i can have an EQ on and you better believe its becoming audiophile grade...)
anyways
yeah, there are 24-bit dacs, but like everyone is saying, theres A) no source material with that much dynamic range (if its in 24-bit, play it back as such, but its dynamic range can't really use all of 24-bit) and B) theres no equipment that can utilize that full dynamic range
its kind of like true 32-bit color for computer displays (this is NOT the same as 48-bit color on your TV, thats a whole different ballpark), yes there is software that can generate ~10-bits per pixel of color information, there is no display device and (for our purposes) no video adapter (Matrox had one a few years back, but it couldn't output 10 bits per pixel, it could only internally process 10 bits per pixel) that can actually render in 10-bits per pixel, even though 32-bit color provides a higher quality image than 24-bit (well, if you have alpha disabled on both, alpha channel does a lot for you in 3D though, but enough about that, back to sound)
basically the easiest way to think about 24-bit over 16-bit, is like 48-bit on your TV, what that 48-bit means is you've still got a 24-bit display (meaning 16,777,216 colors are possible at a given time) but it runs from a 48-bit (i dont know the # exactly, 2^48, its gone plaid) look up table, so you have a more dynamic color selection (its like having half-tones on printing)
basically with 24-bit audio its a similar (not exactly the same) comparison to 16-bit audio, you have more range to work within, but its impossible to actually use that whole range
now, if you're playing your 16/44.1 mp3s back in foobar with the ASIO4All and upsample forcing it into 24/96, thats software trickery, and its not really going to improve anything (unless you have a soundcard with native ASIO drivers, like an M-Audio, in which case its going to improve things because of ASIO, not because of the upsample (and if you could run ASIO at 16/44.1 i'd suggest it))
so yeah, in summation, play stuff @ native if you can, and 24-bit does entirely exist, but theres no physical way to utilize it 100%