If you take a 16/44 master and upsample it to 24/96 all you're doing is making the file larger. As soon as that file is at 16/44 it has already lost all the data. The difference between 16bit and 24bit is simply down to the ammount of seperate volume levels available. At 16bit you have 65,536 different possible levels of volume, at 24bit you have 16,777,216 possible levels of volume.The problem with older digital recordings is that they were recorded at 16/44 or 16/48 and for that reason they will never be better than that, not that there is anything wrong with 16/44 you get on a CD for most modern music anyway. When listening to highly compressed music such as modern rock, pop, metal or dance then like I said I would be hard pushed to believe anybody could tell the difference.
Taking into account higher frequencies like up at 96khz. 44khz is all that is needed to properly cover the frequency range of our hearing (20hz-20khz). Consider that by time we hit our 20s our hearing is already damaged enough that we cannot hear up to 20khz anymore. But even more important is the way music is recorded.
You look at the way most music is EQ'd and you will see that most of the top end is actually chopped off on individual parts, a mixing engineer is trying to create space in a mix to allow each part to sit comfortably and breath so it is common for anything not deemed necessary to be chopped away. That means perhaps cutting from 16K upwards from the guitars aswell as anything below 300hz completely removed, simular cuts commonly present on the vocals, you'd probably leave the cymbals on a live drumkit but some people will cut from those too, any electronic drum sounds will generally sit in very tight EQ pockets . Generally speaking all these sounds sit well within the frequency range between 20hz and 20khz, so having the quality set so high is often overkill, since even taking away the fact that we can't physically hear them it is common for them to be physically removed from the audio mix.
Old music recorded analogue and cut to a Vinyl has a certain charm about it and you can really feel (rather than hear) the dynamics, because there are no set ammount of volume levels, it is infinite and it is organic but it is also imperfect. A 24bit recording is pretty damn close to providing that, probably close enough that we could not tell the difference, infact almost certaintly. The issue though lies with the nature of modern music and the way it is recorded, the nature of digital recording itself, and the hardware we use to listen to it. Ultimately the quality of the music is much more down to the individual recordings, the mixing and the mastering, rather than whether it is 16bit or 24bit.
It's hard to really put it into words, old analogue recordings on Vinyl have a certain charm about them, though they can be noisy they do give a listening experience that is hard to really understand without experiencing it, probably due to the way that the volume levels are so organic, the dynamics are so much more noticeable and there is no "hard" clipping going on, the clipping is more of a soft warmth rather than the harsh sound of clipping on a digital record. Most modern music (almost all of which is digital) is pushed so hard that the job of the mastering is often about how best to hide the clipping and produce the loudest sound that will sound good across many systems, at which point any noticeable differences you would hear in the recording that 16bit vs 24bit could highlight are essencially squeezed out.
Bit of a ramble, I just meant to say that any record that started life as a 16bit is going to forever be a 16bit regardless if you upscale it, it now has the 24bit available volume level differences only that it is still sticking to the exact 16bit template, the data to make use of 24bit is missing. For reference I have a song I recorded/mastered at 24/44 and it is 76mb for 5mins, vs around 50mb for the 16/44 downscaled master.
.Like the difference between a standard def tv and full HD.
