I've made a read-heavy benchmark on a replica set with 1 master and 3 slaves. The client is using the java driver and set the slaveOk(), but the repartition of the requests was unequal.

The loads of the servers were:

• 0.02 for the master (I think the client send it only the very few writes)
• 60 for the first slave (it was the goal of the bench to push MongoDB to its limits)
• 5 on the second slave
• only 2 on the last slave.

I was expecting that the load on the 3 slaves was similar.