At what focus setting do DXO measure the T setting?
Do you accept that as a lens focuses the mathematical corelation between physical focal length and aperture diaphragm pupil diameter changes?
No, I don't know where they focus wneh they measure. Yes, it's true that focusing changes the apparent focal length of the lens (well, not all lenses, but a lack of focus breathing is one reason cine lenses are so expensive). For example, the 100L IS Macro, when focused at 1:1, is giving the equivelent FoV of a 65mm lens. But convention is to state the focal length when focused at infinity, and I would guess that's where DxO tests the T-stop.
Regardless, the vast majority of DxOMark's lens T-stop measurements come in at or lower than (in terms of light transmission, so a higher number) the rated f/stop of the tested lens.
HOWEVER, I think the actual point of contention is that people are too keen to tear apart DXO's musings and any Canon new product, when it's demonstrable that they don't have a clue about the difference between F & T stops, which are distinct, and what the review is referring to.
While I agree with you, I will point out that DxOMark tested the 28mm f/2.8 IS on 17 bodies - 15 of them show a T-stop of 2.8, one (the 1D X) shows T2.9, and one (the 550D) shows T2.7. The 40mm f/2.8 also shows a T-stop of 2.7 on the 550D, and likewise, it's T2.8 on the other tested bodies. With those very few exceptions, every lens on every body that I looked at shows a T-stop that is equal to or less than (again, in terms of light transmission) the lens' specified f/number (or for zooms, the average f/number across the zoom range).
So...how would you explain DxOMark's apparent outliers for a couple of f/2.8 lenses tested on one type of body? A DxO mistake? Defiance of the laws of physics? Or maybe they have a magic 550D body that just sees the brighter side of life?