The shots are excellent.
What I wonder, though, is if the environments really are very taxing to an averaging metering mode.
I mean, when I look at AgentMedia's shot, it appears to me that the overall distribution of light in the resulting shot is pretty much medium gray. The problem, of course, is that I'm looking at a picture, not at the actual scene that was shot.
It would be interesting to perform an experiment: shoot an equally challenging scene with the 40D and another, earlier model (such as a 10D), both in evaluative mode, and compare the resulting images to see if there really has been an improvement to evaluative metering.
Regardless, it's tough to argue with the results you both got. 