Comment # 6 on bug 1186825 from Neil Brown

Thanks for the report.  Sorry for the delay.

Assuming you can make it happen again, could you please run

  rpcdebug -m rpc -s all ; sleep 1 ; rpcdebug -m rpc -c all

while the processes are spinning, and report the output produced in the kernel
log?

My guess from what you have provided so far is that a "connect" attempt fails
in a way that causes the rpc layer to retry the connection immediately. 
Hopefully the logs produced by the above will confirm or deny, and give me more
details.