It has dawned on me recently that the method described above is only good for small data sets. If your code exceeds 30000 records or so it really starts to slow down. Otherwise, sub-reports should be used instead.