We noticed a performance issue on our primary node during the 95 / 5 YCSB in cache test. 
Sometimes we see a spike in write latency on the primary node that is associated with an increased write concern wait time. 
The write concern wait time goes up because both secondaries start checkpointing at the same time, and become IO bound. 

This ticket is for a feature request to stagger secondary checkpointing such that only 1 secondary is checkpointing at any given time, preventing both secondaries from being IO bound at the same time and causing a delay in satisfying write concern. 

See this comment for some background.