I've only seen this occur once before for one subject andI don't know why it occurred.
It turns out the the contrast vector SPM.xCon.c does not get changed (please double check this in your case) and the fields that do get changed do not impact the contrast or T-test values. While its good to be concerned about odd warning messages like this one, this warning doesn't seem to impact the processing streams.
Perhaps one of the SPM authors could provide more details on why this happens for some models.