This should be an interesting subject, because quantifying and understanding the sensor performance is the starting point to getting the best from it. Unfortunately, too many people here are incapable of contributing unless any metric shows that their purchase/favourite company is shown to be the best.
Oh please. I don't see a single Canon user here denying that there's some DR advantage to Exmor sensors. The question is how much, and how much difference does it make in the real world.
If you really think it is impossible then you should post the mathematical analysis that shows it to be so
No. Theory bends to observation, never the other way around. I think I posted this in another thread, so I'll post it again here: try drum scanning a 4x5 frame of Velvia, a 6 stop film, and then down sampling it to 8 MP, which is the DxO normalization. Tell us if 3 more stops of shadow detail magically appear, which is what DxO's formula predicts.
My prediction from years of scanning film: you will end up with a 6 stop, 8 MP file.
The problem is in the definition of DR. You're using theories that are only concerned with white and black points. But photographers are interested in usable photographic detail.
Down sampling may reduce noise and therefore make your blacks blacker. But it doesn't magically open up shadows and produce details that were never there.
I will concede that down sampling can reduce noise thereby making a print of already existing detail
acceptable, where if the noise were still there you might clip levels to black and discard the noise and detail. But it doesn't produce detail where there is none. It doesn't magically allow a 14-bit pipeline to yield more than 14 stops of real photographic detail. It won't even get it to 14 stops because in the real world ADC pipelines are not perfectly efficient.
BTW - Imaging Resource measured the D800 to 13.3 stops vs the 5D3 at 12.5. I trust their methodology a whole heck of a lot more than DxO's.