I\'ve got a MongoDB collection that holds about 100M documents.
The documents basically look like this:
_id : ObjectId(\"asd1234567890\") _re
I would try setting a non-unique index on _reference_2_id, because at the moment, I suspect you'll be doing the equivalent of a full table scan as even though the indexes contain _reference_2_id, they won't be used (see here).
_reference_2_id