Looking back at the change history, there was a change made on 13 Oct 2011 that affected the way the covariate matrix squared inverse was calculated. The diagonal elements of this matrix appear in the denominator of the t statistic, and so under some circumstances this could change the t-statistic -- but it would not affect the numerator, which is the 'overlay' value to which you allude.
I believe the later version is correct, which is why the change was made. My actual memory of this change is very faded now, so I can't say what event prompted it.
