My understanding is that you should use count(rowid), since there is less data dictionary work to do the identification of what columns the * actually represents.