Is it possible that prod has sincedb_clean_after set to a lower value?
To avoid repeatedly reading very old files the ignore_older option might help.
You may be able to filter out old events using something like this. Although I doubt you can prevent the previous day's file being immediately re-read just after rotation.