job in R state but not running and can't be deleted from queue

Hi,

I observed the following behavior today, on an otherwise stable setup 
based on torque-2.0.0p2 and maui-3.2.6p13 (I know, I have to upgrade).

- some jobs being reported by qstat to be in R state with an exec_host 
defined, were found to have no corresponding processes on the respective 
nodes
- the nodes did have the usual files in the spool/, aux/ and 
mom_prov/jobs/ subdirectories
- a qdel on the jobs, came back with exit-status=0, but the job status 
did not change
- pbs_server logs showed that SIGTERM and later SIGKILL were sent to the 
job, but the usual final message with the job's exit-status and used 
resources was found absent
- in the mom logs, there were messages for job start, 
"scan_for_terminated" and "Terminated" for each job. But the job state 
was still R

- other jobs on the same nodes worked fine during this time

- restarting mom on the nodes helped - the jobs could then be qdel-ed

Has anyone seen this behavior before? Are there any other patterns I 
should be looking for? Is there a better/cleaner way in such cases to 
clear the queues?