In a sharded cluster, aggregate can do a lot of work inside the mongos process. Heretofore, mongos has had a modest computational load; aggregate changes that. We need to make this explicit in the docs someplace.