I see the difference in noise and it sure is there! But, this example is flawed in my opinion since the exposure in these photo's was just wrong. I have seen that current Sony sensors offer better shadow recovery, no doubt about that. But in this particular situation it was not necessary to have such an underexposed image and the same photograph could be achieved by both Canon and Nikon camera when properly exposed.
But in another scenario, this exposure could have been needed to avoid clipped highlights. So what you're really complaining about is that I didn't - in my infinite free time during which I happen to have both a D800 & 5DIII on my hands - find a scene with enough DR. That's a fair point, sure, but also - as you can hopefully see - a bit unfair. Because the end result would be the same.
And, frankly, that's what I'd say to everyone here who, in all the comments that followed my original post, indicated that my shot was underexposed. Meanwhile, just a few minutes before this scene, the sun was high enough in the sky that the sky would have blown. So are you complaining that I didn't shoot just a few minutes earlier when there was more scene DR? Would that have changed the end result? Are you arguing that there do not exist scenes that require more DR than what Canon sensors can provide?
Furthermore, according to my histogram when I was shooting, my channels were clipped in the sky! Therefore, the camera led me to believe that I did accurately 'expose to the right and for highlights'. So this is the result an informed photographer would have gotten anyway. Now, if you're going to complain that I didn't have a computer with me to actually check the RAW values as I was shooting, then...
There are many valid suggestions you could make. *Perfectly* nailing exposure within 1/3 EV to get your highlights just short of clipping is not one of them (though it is a noble effort). Not the least because most cameras don't accurately tell you when you're just short of clipping in RAW - in my experience, they tell you you're clipped well before you're actually clipped.
Now, you could have told me to bracket, or bracket for HDR. Valid points - but sometimes that's not an option for certain types of photography. My point is: the conclusion doesn't change. D800/Sony sensor performance gives you much more DR and much more processing latitude. That's useful to some people in certain scenarios.
Also from http://www.luminous-landscape.com/essays/on_safari.shtml#suckout :QuoteThe bottom line is that the vast majority of the tonal information that a sensor can capture lies on the far right side. In fact, usually 75% of the total tonal information a sensor can capture lies in the small right hand area of the two top F/stops just below pure white. And yet, almost all cameras leave the factory calibrated to center the histogram instead of moving it as far to the right as possible. To make matters worse, the screens in the backs of cameras are also calibrated to show a good exposure with the histogram centered. This is nonsense!
So what are the consequences of Tonal Suckout? As the name implies, it is as if someone grabbed your image and sucked the tonality out of it, If there are millions of shades of green in a landscape, you may end up with a few dozen. If there is a terrific richness of tonalities in a face, you will end up with a small sub-set that makes the face look bland.
Maybe useful to keep in mind the next time you want to get an underexposed image!
Just FYI, that's actually a fallacy. Read Emil Martinec's extremely informative treatise on noise: http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p3.html#bitdepth
Essentially, any extra 'tones' will oversample shot (statistical) noise. Consider a signal of 200 photons generated ~100 e- of signal (QE = 0.5). At ISO 100, say we bin 4 electrons 1 DN (digital number, or increment, in the digital file). So that signal is recorded as 25 in the RAW file - pretty darn low (dark). Now, b/c of shot noise, that signal already has a stdev of 10 e- (sqrt of 100), or 2.5 DN. So there you go, you have enough bit depth to still sample the noise in the signal. So it's not limited bit depth/tonality that's the problem - it's the amount of light itself. You expose to the right (ETTR) to minimize statistical/shot noise.
Or to avoid running into the noise floor of Canon sensors... make sense?