Despite not knowing anything about your dataset and insertion/deletion patterns, you may want to consider power of two allocation. It's a new feature in 2.2 that costs some additional disk space up front (new documents potentially get 2x padding), but can save a lot of disk over the long run by eliminating many data access patterns that add to fragmentation. One (pathological) benchmark I ran saved 800x the disk space after 1,000K (edited from 100K), insertions and deletes of 1-10KB documents.

http://docs.mongodb.org/manual/reference/commands/#usePowerO...