Cluster Management

All the nodes should have the same configuration, so all of them should have the same performance. If this is not the case, probably 
1- You have the data skewed ( you can check projection_storage table to see that)
2- You are not using a load balancer and all your queries go to same initiator ( you can check query_profiles and check the node_name column and see if the data is distributed)
Hope this helps, 
Eugenia