I found quite an interesting article by Mason Resnick, formerly Popular Photography Managing Editor and now an Editor in Adorama Learning Center.
Cutting the long story short, Mason is sure that all 6 currently available full framers display incorrect ISO, with 1/4 up to 1/2 (!) step underexposure.

I tried to summarize his lab results in the table here

The thing that I especially loved was the test of Nikon D3X @ ISO 50 which led to the conclusion that ISO 50 at D3X is in fact an in-camera noise-reduction post-processing, i.e. "not a real thing"

For APS-c size sensor bodies Canon 40D was measured with 1/3 underexposure for all ISO speeds. Should I apply a permanent exposure compensation for my 40D now?

I normally have my 40d at 1/3 or 2/3 over, so it seems he may be right

Its the one thing I agree with DxOMark about.

All the cameras have different results when you put the ISO to 100, put them same lens on all of them and shoot the same picture with the same shutter speed and aperture.  This can only be to the misrepresentation of the actual speed of the sensor.

Its crap, because we all know that ISO is a standard, but its just a way for the marketers to market their 6400 ISO and other crap.

There is nothing you can do about it but compensate.  Unfortunate

The only thing I believe is the histogram.

I think that the histogram show data for the internal camera processed embedded preview rather than for actual raw data. That used to be typical I think. Result being that we can not believe the histogram displayed on the back of the camera because of the internal processing applied to make the preview. This issue is very well documented on camera tecknical forums. Problem may have been addressed in later models. In the end you can only trust your spidey-senses.