Have you tried dropping into the node shell and running a systat -M?
(node run <node-name>; priv set diag; sysstat -M)
Dump the output to something where the text doesn't wrap so you can line up the data fields with the columns. It might help you zero in on the part of HostOS that is chewing through CPU.
In the end, you could also re-open your case and do a perfstat gather. The new perfstat tool is exceedingly easy to use compared to the CLI version and we were able to sleuth out our odd performance behavior case a while back with the assistance of someone on the performance team.
Good luck!
Chris