A bit more on this.  Looking a the time when my job really was dragging, I see a high percentage of CPU time going to "iowait" which I assume has to do with S3 and disk latency.  See the yellow areas below in the CPU utilization graph.  I am reading from a common s3 place into my own region and instance.  So, am taking steps to isolate myself from AWS public.  So, could be file size, could be AWS latency issues: