The mongos currently does very basic round robin query distribution across replica secondaries. Due to degraded hardware, we sometimes have replicas which answer queries much slower than other replicas, and ideally the mongos should that and balance appropriately.