We've found in the common case it's better to have fewer pods, but rather have nodes that have more cores to squeeze out more throughput. This is so less work is delegated over the network and so that the Go runtime has more cores to process things in parallel.