yes, I do experience similar underexposures, but it in my case it depends on the lens e.g
24-70 2.8 L mkI -2/3 EV
70-200 2.8 L mkII 0 EV
Macro 100 mkII +1/3 EV
85 F1.2 -4/3 EV
35 f2.0 mk II 0 EV
of course these are rough averages
Most were shot in AV mode, around F5.6, evaluative metering.
It would be great if we would have something similar to afma, so an exposure correction by lens ?
During some tests I have found that the 5d3 and 5D2 underexposes in general by -2/3 stop. Checked with sekonic and other camera (D300). Moreover I observed similar behavior as mentioned above, so it depends somehow on lens:
70-200: 0 or -1/3 EV
85 1.8: 0 or even +1/3 (i would rather say it is perfect)
50 1.4: -2/3EV
135L: -2/3 EV
One more thing which could be strange is that a lot depends on light environment. When I shoot outdoors exposure seems to be much more accurate than in indoors. Almost all indoor shoots are completely underexposed which seems to be strange behavior as a metering system should meter everything as a neutral gray giving histogram peak on the center. Mine is almost -2/3 EV to the left. Once upon I time I run some test with 5d2, 5d3, Sekonic light meter and Nikon D300. I checked metering indoor on neutral even surfaces like white or gray wall. Only canon cameras results were -2/3 stops different from meters. I've sent it to service but they stated all was in line with standards. I spoke with some other users and some of them told me that the use a +2/3 correction all the time. It is a kind of a problem as shadow and dark parts of photos are hard to push with canon's sensors so the tendency should be completely different.