Well, to summarize:
Backside Illuminated Sensors (BSI) only offers significant benefits when your pixels are ultra tiny...say 2 microns in size or less (many point and shoot/phone camera sensors have pixels as small as 1.9 microns, possibly even smaller these days, what with 40mp+ phone cameras floating around...!!) For comparison, a 7D has 4.3 micron pixels, the D800 has 4.6 micron pixels, the 5D II has 6.4 micron pixels and the 1D X has 6.95 micron pixels. Readout wiring is in the range of hundreds nanometers (fraction of a micron), so it isn't usually a problem until your pixels are around 2000 nanometers or less (where a couple hundred nanometers is a significant percentage of your pixel area).
Sony Exmor mondo badass hardware noise removal.
Canon uber suckass hardware noise removal (well, ok..SO-SO mediocre hardware noise removal...to be fair I am a Canon user after all. )
Why do you think that Nikon D800's Exmor sensor settled at 36 Mpix?
Wouldn't be a much better solution WRT noise removal a, let's say, 22-24 Mpix sensor? Now I'm thinking that if Nikon would had a 18-24Mpix Exmor then the 5D3 would be in serious trouble, because Exmor's NR hardware correlated with a rather big pixel size would have an amazing output even at high ISOs.
What do you think?
I don't really think there would be that much of a difference, honestly. Pixel size has more to do with QE, or Quantum Efficiency, than with noise. Improvements to QE have the side effect of improving noise characteristics, however outside of the lowest ISO settings (and even then, only to a slightly lesser degree), the very vast bulk of "noise" is photon shot noise, rather than electronic forms of noise. Photon shot noise is a matter of physics, due to the random and otherwise unpredictable nature of light packets (photons.)
All things being equal, once you eliminate most electronic forms of noise (such as is the case with Exmor sensors), more pixels is always a good thing. Higher spatial resolution at that point is always better than lower resolution, regardless of diffraction. At some point you reach a spatial resolution where even the finest details that you could possibly photograph require representation by many pixels (i.e. the smallest aspect of detail requires at lest 10x10 pixels to be represented.) Assuming little or no electronic noise, there is a HUGE benefit to that, because now noise is a SUB-DETAIL level issue, and noise removal is now largely and intrinsically a detail-neutral process.