ES Server Utilization

I have some seriously powerful servers which appear completely under-utilized at this point. I'm looking for suggestions on how to increase their performance in order to increase indexing & querying efficiency.

I have a 4 physical node cluster -- each device has between 32 and 64 processors, about 256g of RAM (~128g committed to ES) and 20T of storage. I am running ES version 2.3.4.

The issue is that these servers are using hardly any CPU, and out of the 128g RAM available, the cluster uses only about 20g. Is there a way to commit more resources from these servers? I'm considering running multiple ES nodes on each machine, but after having combed through the forums it appears that's not an ideal solution.. Any ideas?