We have a 5 node replica set, running 2 MMAP, 2 WiredTiger and 1 arbiter instance. Upon making WiredTiger a primary our web-server become unresponsive. Further investigation revealed the number of connections had gone from ~25 to 120.

The problem appears to manifest itself when our system is on catch up and thus more heavily loaded that usual. The load is not exceptional though, with only a single threaded service performing multiple reads and writes to MongoDB, and a web-server connecting to MongoDB and performing reads for the most part with only a handful of users.

The MMAP instances do not appear to exhibit this problem.