Hmm...are those graphs for one beam or for all the beams that were collected for the final result?
I do agree though that the statistical analysis is the most likely place for an error followed by the GPS clock synchronization.
The next few weeks should happily settle the former.