Consider a really small pixel that can absorb only 1000 photons to maximum well capacity. Square root of a thousand is 32. Photonic noise is 3.2% of the signal. A big difference.
I'm a little confused. Is 3% relevant? With one single f-stop being 100% more light (photons) or 50% less light (photons) is 3% meaningful?
I'm of the sense, that the future is at the pixel level. From photographs put out by the companies, the light collecting area is a small percentage of the pixel area. Is there a future technology that can improve this? I'll bet there is.
Also, can less noisy amplication allow the pixel to perform well at lower light levels. This permits the opportunity of reducing pixel size with the same performance. Did Sony come up with a breakthrough and share it with Nikon. Are Fuji's R pixels (it may be the other one- well anyway the small one) clean enough to be used as the principle light collecting device. maybe not now but coming soon.
The hard to grasp concept is the pixel is only a data sample, vs. film which is analog and continuous capture. Lots of dead space between pixel capture area (silicon).
So if we're reassempling and interprelating data points how does chip (vs. film) size matter anymore. I don't see the relationship between size and output as we had with film where the enlargement reduction of the larger negative was the principle advantage of going large. Tri-X was Tri-X whether 35MM or MF or 4x5. But the final print looked distinctly different when printed at the same size.
Software improvements can work with the data to improve color and resolution. Oversampling worked with digital music, is there an equivalent for the visual arts. Probably.
So, to me, it means, I must think in terms of good data. This leads me to the conclusion that pixel quality supported by lens quality are the two principle factors determining our future.
I bought a Nikon Coolpix 880 (7 megapixel I think) for a walking about camera. I have a shot from northern NM that looks the full equal of my 20D and D70. I did put it on a tripod, used the lens better aperature, had pretty light, and the subject was pleasing. Little tiny pixels, noisy for sure as higher ISO, crappy electronic viewfinder, but in a narrow element it worked as well as anything at that pixel quantity. Now no one uses it that way, the vast majority of shooters (including the spouse unit) use this camera P&S (why I like the new anti-shake) as do my digital artists in reference gathering.
Back to the point, Its now all about collecting data and chip area is no longer the critical determinant as it was in film. I think long term LF is dead, and that MF will become view cameras (aka sinar F3) for the movements that the view camera allows. hand held cameras over 10 mpxl's are capable of doing most everything other than giant prints and even then for some subjects do fine.