@ Bosman, you sir are an optimist amongst sceptics I fully agree with you that the current 7D has filled an important niche in the market, particularly as a 2nd-body DSLR, but I (anyway) atrribute that to the way in which Canon developed the 7D by asking professional photographers (the infamous 5,000 questionnaires) what it was they wanted in a crop-sensor camera, resulting in a 'poor man's 1DIV' (apologies for such a drab analogy), yet your optimism foresees a kind and benevolent Canon Inc. making a new 7D2 the 'poor man's 1DX'. I just hope that you are indeed correct and that Canon produce a 7D2 with unique AF + Speed, whilst improving DR + Noise, with the added constraint of possibly even smaller pixels.
Canon sensors have yet to push past the 50% Q.E. barrier, on any sensor. Sony sensors, and a few Nikon sensors, have pushed past the 50% Q.E. barrier, and Exmor is pushing 60%. Improvements to Q.E. can improve photon noise, even with a smaller pixel.
Apparently Sony have improved the tech in their new Alpha-99, which uses the same 24.3MP Exmor sensor as the latest Nikon offerings, but their 'shallower' pixel wells allow more light to hit the photo receptors which are also a bit larger ( I think of this improvement in micro architecture design as something akin to what Intel achieved with the Sandybridge & Ivybridge chips i.e. 32nm and 22nm chip thickness as opposed to older 45nm+ chips). Anyway, if Canon can produce a 7D2 that shoots 10fps and improves ISO by at least 1 stop, and has an AF system at least as good as the classic 7D, then I'm sure it will sell in large quantities.
Same thing here...your talking about Q.E., or quantum efficiency. There is a certain amount of light lost at each pixel, due to a variety of things. Some light is lost when it reflects off of the filter stack just above the sensor. I wouldn't say the IR cut/low pass filter stack is a significant source of light loss, but is a key factor. The color filter at each pixel absorbs a fairly significant amount of light...you lose at least 15-20% here, if not more...depends on how strong the filter is. Photons that strike parts of the sensor die that are not the photodiode may reflect out of the pixel well, or convert to heat. Not every photon actually converts to an electron. By minimizing these losses, you improve the quantum efficiency of your sensor.
Assuming you had a FF camera with 30% Q.E., and an APS-C camera with half the pixel area and 60% Q.E., there is no reason the APS-C camera couldn't perform just as well as the FF camera from a noise standpoint. Technically speaking, you could capture the same number of photons per pixel despite the fact that one pixel is 1.4x smaller. Canon sensors have around 45% Q.E. today, so unless they can improve Q.E. to around 90% for the next 7D, it wouldn't necessarily perform as well as a 1D X...but neither would it perform 2x as bad. An APS-C sensor with 60-65% Q.E. could perform VERY well on the noise front. As for DR, that is a whole other beast, as it is limited on the shadow end by electronic noise in the sensor (i.e. noise caused by dark current and differences in response from pixel to pixel, etc.) Canon would need to implement some fairly radical fabrication improvements (which I believe they are indeed capable of), and possibly a shift to some kind of CP-ADC with digital noise reduction (vs. analog noise reduction).