I asked about that same question early in regards to the differences between the data reported on MSAR, schools' websites, and us news.
As far as I can tell, there are multiple reasons for the MSAR data being different:
1. MSAR data uses accepted students
2. MSAR rounds up numbers for the MCAT and GPA
2. Some schools' websites use historical class avgs(I know Princeton Review does it with their data and I think US News use to)
3. Im guessing most schools probably have a 2 hump GPA/MCAT distribution graph for their accepted candidates because of the usage of nonacademic factprs in the decision process such as legacy status, connections, disadvantaged/URM status, and etc. Most accepted students are probably skewed towards the high stat end(based on the data seen on mdapplicants.com), but a small minority with relatively lower stats bring down the mean such that there ends up being a noticeable difference between the median and mean.