From one of my economics Hill guys:

Jonah, I thought I might shed some light on your question about possible trends in the size and direction of revisions to GDP data. According to a 2004 study by the Bureau of Economic Analysis of GDP revisions from 1983 through 2004, “none of the mean revisions in GDP are statistically different from zero.” While it may appear to the naked eye that revisions are generally in a positive direction, that is not the case.