I'm not convinced that Canon's DR is a result of worse sensor technology and not the result of a deliberate design objective to limit it. There's nothing, AFAIK, in CMOS or CCD sensors fundamentally that would produce the characteristic sloping curve you see from Canon's cameras. Moreover, virtually no other camera manufacturer shares the same kind of curve or the pervasiveness Canon has when you look at DXO's DR charts. In fact the only Nikon cameras that share that pattern that I'm aware of are the D3, and it's derivatives the D3s and D700.
So Far as I know, DXO, contrary to their claims, doesn't measure just the sensor, their measurements cover the whole signal processing chain from sensor though amps to ADCs, to whatever the processor does to the values before outputting it as a RAW file. While this is certainly is more for useful for the images you'll see but doesn't really provide a means to say anything about the sensor tech.
So why is Canon doing this? That's a darn good question.
I think Canon and Sony/Nikon have made different trade offs in their sensor design.
The Sony Exmor technology has the analog to digital converters (ADC) on the sensor chip. The Canon sensors take the analog signals off the sensor chip to an ADC in another chip. Very weak analog signals can pick up noise from the circuit boards. This is why the readout noise of the Canon sensors is higher than the Sony Exmor sensors and why the Sony Exmor sensors have better DR and lower deep shadow noise at low ISO.
So why doesn't Canon put the ADC on the sensor chip? There may be some patent issues, but I suspect the real reason is that the Sony Exmor technology has problems with video. The ADC on the sensor chip generate a lot of heat when the sensor is being read quickly, as in high frame rate video. Some Sony DSLR-like cameras that used Exmor have had problems with sensor over heating. Sony has fixed those issues in it most recent cameras. The fix probably involved better cooling for the sensor chip.
All FF DSLR video cameras face the problem that they cannot read all the sensor pixels at 30 fps to do HD video. Nikon is using some kind of pixel skipping to reduce the amount of data that has to be read off the sensor. This means the D800 is not using the whole sensor area, even the whole area within the HD image, to generate its video signal.
Canon has developed on sensor circuits that allow the analog signals of multiple pixels to be mixed for downs sampling to HD video resolution, so they only need to put these down sampled HD video pixels through their off chip ADC at 30 fps. Canon uses analog signals from the whole sensor area, at least the part that is in the HD aspect image. This allows the Canon sensors to give better high ISO video.
Dan Chung from DPR said that up to ISO 1600 the D800 and 5DIII video noise was similar, but above 1600 the 5DIII clearly had lower noise.
Sony Exmor puts the ADC on the sensor for lower read noise and improved DR. Canon puts analog video down sampling circuits on the sensor for improved high ISO video.
I think Canon thinks that the DR they have now is good enough, and it is for most (but not all) photographic applications. They don't want to trade off the video performance, especially the high ISO video to get better DR.
