After investigating further it seems to not be related to when a full backup is taken. It seems to be something to do with either the index optimisation scripts we run, or the fact that the Sharepoint workflows seem to insert a lot of rows into the EventCache table between midnight and 2am.