Been following this thread, and want to point out a couple of things. First, absolutely, mastering in 24 bit will allow far more freedom from having to ride the level controls to run as close to 0 as possible during capture, allowing for lower levels that can be boosted in post without introducing noise. However, a 16 bit recording with optimized levels will not sound much different than a 16 bit product produced from a 24 bit master using the same front end (mics>pre). The reason that UV22 and SBM sound better than straight 16 bit A/D is precisely the fact that they quantitize at 24 bits, and then use a noise shaping filter to remove the digital noise that the very act of quantitization creates. This makes the actual perceived dynamic range to approach 18-19 bit depth to the listener.
The dynamic range of a rock show through a PA is about 40db, and a jazz show, maybe 60db. Watch your levels during a show. Do they sweep constantly from far below -12 up to -2db? I have rarely seen that, except for a single acoustic instrument recorded in a pin drop quite setting. So, if your levels range from -12 during the quite portions, and hit-2 during the loud portions, that's only 10db of dynamic range.
Using the example of the average listening space, cars driving by, lawn mowers, dogs barking and/or kids playing outside, and HVAC systems inside, it's hard to imagine the average Joe sitting in an acoustically dead lab setting listening for differences in dynamic range between 24 and 16 bit. To me, the real advantage of 24 bit is the ability to simply not be as concerned with managing the recorder in the field to optimize levels. I am not saying you don't have to be a "good" a recordist with 24 bit, but you definately do not have to be as good at setting and controling gain live as you do with 16 bit.
Sampling is another very misunderstood thing. Regardless if it's PCM or DSD, digital samples are taken 2 per frequency per second, one for the left channel and one for the right. It goes back to basic electronic theory of hertz measurement of cycles. 48khz takes 2 samples of each frequency per second from DC all the way up to 24khz, at which point the anti alaising filter cuts off the analog input. 96khz takes the same 2 samples per frequency per second from DC all the way up to 48khz, far beyond the ability for 99% of capture or playback systems to reproduce, and no human can hear. There are more "points", but these are not compressed into the same audible range, as with the difference between standard and high def video which has more actual lines of resolution within the same screen area.
DSD does the same thing, but takes 2 samples per frequency per second into the 2.8ghz range, using 1 bit per sample, and because the samples from DC to 24khz are represented with less bit depth than PCM, results in the industry having mixed opinions as to the advantages of DSD, and why PCM was not replaced by DSD outright, which would have happened if the opinions were not mixed.
The reason that any higher sampling frequencies above 44.1 sound better at the capture point is due to analog filters to prevent crossing the Nylquist Frequency. To prevent a signal higher than 22.05 khz from hitting the A/D, a filter starts to act on the signal at 20khz, because there is no such thing as a perfect brick wall high pass filter, and it needs 2khz of roll off to kill the input by 22.05 khz. This roll off starting at 20khz is audibly noticable. Recording at 48khz eases the task of the filter, as it does not kick in until 22 khz to roll off to full attenuation by 24 khz. Recording at 48khz or higher, and resampling in post does not have the same impact as the filtering is digital, so the theoretical upper frequncy reproduction limit of 44.1 can be achieved, which is 22.05Khz.
So, I would answer these questions this way:
Is 24 bit better than 16 bit? Well, it depends on the source, recording environment, capture front end, and how much attention you want to spend riding the gain controls of your recorder.
Are samples from 48khz and above better than 44.1 at capture? Yes, but above 48khz is probably unecessary, but does no harm other than taking up more storage space.
Remember, digital recording is 2 samples per frequency per second, with PCM using more bits per sample to represent dynamic range and that is all. Master at 24 bit 48khz or above, use Wavelab UV22 to dither to 16/44.1, and that will sound better than a 16/44.1 master. Or, use an AD1000 or MiniMe, or SBM in the field at 48khz and optimize your levels correctly, and you will end up with the same result.
Sorry about the lengthy post!