In slow query in a sharded cluster would be useful to log stats on the sharding filter stage.
For example the number of skipped orphan document after fetch or number of index key related to orphans that the engine had to go through (if possible)