[Shiraz (Ray)]

A few comments

Most cameras only have about 12 bits of effective resolution. They may manipulate the data so that it covers 16 bits, but there is normally no more than 12 bits of information above the noise. A 12 bit DSLR that provides 12 bits of real data will produce just as many usable shades of grey as a 16 bit system that provides 12 bits of real data. The only chips that can get much better than 12 bits of real data will cost as much as 10 DSLRs.

The dynamic range of chips with small wells is pretty much the same as the dynamic range of chips with big wells, since the big chips have higher read noise. Run the small well chips with more numerous short subs and you get the same dynamic range as large well chips with longer subs - you can't use short subs with big wells because of the high read noise. Images from chips with small wells generally have saturated stars because their users have been convinced that they should use long subs - if they use short subs, the problem goes away.

small pixels do not have more noise than big pixels - in fact they have less noise due to the reduced amount of silicon. Images from full frame DSLRs look less noisy because the sensors are more sensitive due to the bigger pixels intercepting more photons - thus the gain (and noise) can be lower. However, this extra sensitivity comes at the cost of much reduced resolution due to the bigger pixels.

The extra hardware around a CMOS pixel is not a problem - microlenses are used to focus the light from the full pixel size down to the physical pixel, so there is no light lost by collision with non-sensitive hardware. Modern CMOS is more than competitive with CCD technology in all respects.