When using a mongos that can have a stale afterOpTime (not using it to communicate with other nodes), it is crucial that we guarantee that the config server is always fully replicated. The afterOpTime is usually gossiped between the mongos and other nodes, so that any majority read request will have a certain minimum time to read at.

Failing to replicate in lieu of having an updated mongos afterOpTime can allow the mongos to read from a minority config server node that has an out-of-date majority commit point.