During a PR discussion it was brought to light that users of ReplicationCoordinator are trying to work around methods that take the _mutex lock on the belief that it is a heavily contended mutex.

As most of the time the operations performed are read operations (all const methods for example) it could be of interest to explore the usage of a shared_mutex.

The only problem this could cause is with write-starvation. An option to tackle this is to set a preference on Linux via pthread_rwlockattr_setkind_np.