Going from single-node to cluster can be as bad as going from L1-cache to disk.