The wtimeoutMS as a valid number spec test considers that a URI with wtimeoutMS and no w value (default = 1) to be valid. Since wtimeoutMS only pertains to replication, I would consider that specifying it when w is <= 1 is a logical error.

I noticed this while looking at aefdd2b for CDRIVER-2621. libmongoc previously ignored wtimeoutMS in that case, which seemed more correct.

The only benefit I can see with the current behavior is that wtimeoutMS will always be parsed and ensure that the driver creates a non-empty write concern. In the event that the cluster has a default write concern, this will ensure that the client's write concern is sent and applied (even if wtimeoutMS has no effect).