Use a second, less quick upper bound based on character occurrence counts.