Pardon me for reviving this old and contentious thread, but one thing has come to my attention recently that might additionally explain matrix/evaluative metering inconsistencies.
Although it was indeed briefly mentioned in the initial Wikipedia article I referenced to on page 4 (where the main "culprit," was a bank of "30,000 scenes"), I've come across a diagram that explains visually how
focusing points might contribute to differences in exposure of the same scene. In other words, matrix/evaluative metering shifts exposure emphasis based on the selected focusing point. The illustration comes from this
Canon link.While I was aware of such a function (one of my Canons had a menu option to link AE point to AF point) and I was using it actively, it only recently dawned on me that it might have contributed to the issues Justin was facing. As much as this function is predicated on the assumption that where you focus is where your main point of interest is, thus the overall exposure should be based on that too, there are situations where this may not be necessarily so. In other words, although the whole scene is the same, two (or more) subsequent exposures may not be, depending on where you focus.