Are join algorithms like this really the alternative to compound indexes? I understand that they reduce the memory required for joins, but the amount of data involved in a join is the same.