Yes in this example it was a single node cluster but we also tried to scale this to 5- 122GB servers and the CPU problem carried over to each server and the job did not speed up at all.

We tried to also mix up compute instances for divers and delta instances for the workers.