Well... There are a great many (infinite, really) possible input signals that will all generate the same 16/44 recording, but some of those inputs are more likely than others. You could simulate the likeliest and "record" it at 24/192.
It's also "impossible" to upscale an image, but the difference between bilinear and bicubic and sinc upscaling is readily apparent.
This of course assumes that the listener is capable of discerning 24/192 sound.
Dure, but there's no reason to expect you lose anything, either, barring implementation errors. Lots of the TV you watch looks fine, despite being a 720p signal displayed on a 1080p display. Same concept.
