There was a relatively technical article recently referenced in one of the ETTR threads and my recollection is that at some point, at around 12-14 bits, noise renders any "larger container" superfluous, at least with today's AD converters.
You can't simply compare DSLRs and MFDBs by bit depth, or say that anything beyond 12 to 14 bits is meaningless - as with most things in life, it's more complicated than that. People who want to boil this debate down to simple black and white answers are only deluding themselves.
A/Ds come in various quality levels, depending on cost and sampling rate, among other things. You can certainly get very high sampling rate 16-bit A/Ds, suitable for imaging applications, with 2 to 4 bits of noise, i.e. the A/D itself generates 2 to 4 bits of noise, leaving 12 to 14 bits of real, noise-free data (obviously any noise in the signal being sampled by the A/D just gets passed along, along with the additional noise contributed by the A/D itself). These are relatively expensive, however. Lower cost A/Ds typically generate more bits of noise, delivering fewer bits of noise-free data. You can get up to 24 bit A/Ds with more moderate sampling rates with similar noise capabilities, i.e. you get 20 to 22 bits of real, noise free data (by noise free, I mean noise contributed by the A/D itself - any image sensor output will almost certainly have plenty of noise in the lower bits if you are sampling it at 24 bits).
So if you are going to argue the merits of bit depth, the first thing you have to ask is what is the quality of the A/Ds you are using. All 14-bit A/Ds or 16-bit A/Ds are not created equal. I don't have any hard data to back this up, but I'm pretty sure all of the new 14-bit DSLRs are using actual 14-bit A/Ds, which means they are getting, at best, 10 to 12 bits of real, noise-free data, assuming they used expensive low noise A/Ds. Furthermore, I believe that at least some of the MFDBs are using more expensive 16-bit A/Ds (and maybe even higher bit depths), and downsampling to 14-bits (throwing away the noisiest bits), and then saving those 14-bits as 16-bit RGB values (a bit confusing, I know). If true, this means MFDBs are getting 12 to 14 bits of real, noise free data (or more, if they are using high quality higher bit depth A/Ds, and then truncating off the lower bits for storage as 16-bit RGB).
Given the price differential between DSLRs and MFDBs, I also suspect DSLRs are using lower priced, and hence noisier A/Ds, while the more expensive MFDBs can afford to use higher cost, and hence less noisy A/Ds. This is simply guesswork on my part, but it seems reasonable. So again, not all A/Ds are created equal. I find it very plausible that the higher cost MFDBs are using more expensive, lower noise A/Ds in their processing path than the DSLRs are.
Anyway, my point is that making a simplistic argument that 12-14 bits is as high as you can go is just that - simplistic. It certainly doesn't take into account the parts budget that the designer has to work with, and the quality of the parts he is forced to use for cost reasons, or the quality of the parts he can afford to use because he is designing them into a higher priced product. In other words, part of the reason MFDBs are so expensive is that they use higher quality, lower noise components (of course, there are economies of scale, etc, but parts cost is also a factor).
And there are a host of other noise sources that have to be accounted for, in addition to sensor noise and A/D noise. There's amplifier noise, and power circuit noise, and coupling noise. All of these are price sensitive - the higher your parts budget, the lower noise parts you can afford to use. And most of these get worse as temperatures rise, so your camera or back has to have good thermal management and heat dissipation to get the best out of the equipment.
So as I said at the beginning - You can't simply compare items by their bit-depth, or say that 12 to 14 bits is the effective limit of what you can have. It simply isn't so.
