I am tracking down possibly similar behavior on a production server and a colleague introduced me to Dstat. I am using it right now to try and track down the root of my problem. May or may not be helpful in figuring out exactly whats causing this and then on to figuring out to fix it.
http://dag.wieers.com/home-made/dstat/#
then run something like: dstat -t -l --cpu24 --sys --tcp --vm --nocolor --top-io --top-latency-avg --output /tmp/dstat.csv
