The front "mongos" component of the free shared cluster is behaving erratically. We are investigating it.
EDIT: There was sudden drops in free disk space. We change the logging method and it seems to have stabilized the system. We are still working on figuring out the issue.