There are use cases when you don't want all the distinct values (or they don't fit in the limit), you want their number. It would be great to have and optimized command for that (as an alternative to slow Map/Reduce that creates temp collections).