dtaylor: How do you test DR?
Stouffer step wedge and visually inspect the results.
Furthermore, your results match DPReview? DPReview doesn't test RAW dynamic range...
They used to report RAW and JPEG.
My own 'real-world' tests also show ~3EV better DR on the D800 when I do side-by-side shots of high DR sunsets with my 5DIII vs. D800;
I have a hard time believing 3 stops, though I must admit I have not formally tested these bodies.
Put another way: I have to overexpose my 5DIII by 2 to 3 stops at the very least to get its shadows to match the cleanliness of lifted shadows of the D800 file that was underexposed to maintain highlights.
"Match the cleanliness" is a wide open question. Are you matching at 200% in PS or in a 20" print? And what constitutes a "match"? Does the 5D3 not "match" if there's a hint of noise that's irrelevant to 99% of uses? And to what degree does color play a role? (When you push RAW converters you can often recover detail that is correct in terms of tone, but incorrect in terms of color. How much of this you're willing to accept will alter the final judgement on DR.)
So, respectfully, I fail to see how DXOs DR & SNR, etc., numbers are the 'odd ball out'.
When I've compared their results to other sites, or to my own experience, they have not matched. One example: according to DxO the 7D (Canon's 18 MP sensor) has little DR gain over the 10D / 20D. I could tell you before formally testing them that it was large, 2 stops easily.
Now maybe I'm being too harsh. Maybe their current tests are better, or maybe it just so happens that the cameras I compared were the odd balls, not the entire testing methodology. I'll take another look. But DxO seems easily thrown by small factors, or easily gamed. Michael Reichmann was a big fan when they first started, then dropped them later because really tiny things would shove one score well above another, and not just on DR.
To clarify: I don't at all mean this to be a personal attack; just looking for clarification.
As was evident from the tone of your post, and I appreciate that. I'll look more carefully at DxO's latest results.