The config server side of the balancerStop command does not obey maxTimeMS and because of this can remain running even though the mongos side has timed out earlier. For longer-running balancer rounds this can lead to many threads getting stuck on the config server.