The IR method has a lot more variables and they don't test RAW they test after conversion RAW with cooked in NR of various who knows what degrees
NR does not affect total DR. And they use ACR with default settings for ALL cameras.
(and that also explains how they manage to get this type of DR to actually measure higher than engineering DR measurements of the RAW file).
If DxO was measuring what you think they are measuring then this would be impossible no matter what processing was performed.
Go take a long, hard look at jrista's processing of the infamous 5D3 vs. D800 online test. DxO claims that there is a 2.5 stop DR difference between these two cameras. If that were the case then that door and those tiles in the far back should be BLACK. No detail or image at all in those regions, just blocked up shadow.
Instead we see the same features that we see on the D800, just with a lot more color noise.
The color noise impacts our ability to push the shadows, or shadow latitude. But the DR is darn near the same. It certainly is not 2.5 stops less.
There is your direct, observational evidence that DxO is wrong.
Perhaps more importantly for someone buying a camera, the final result with NR shows just how small the difference ends up being in the real world. Yes, the D800 is better. Could you spot it on a 36" print? Probably, but it certainly would not ruin the print. 24"? Most people could not without being told to look for it up close. 12"? Nope.
How often do you shoot a scene intended for 36" prints yet blow the exposure so badly that you need to push shadows this hard?