An instance of QZSRCGI won't end after ENDTCPSVR

We host an e-commerce website which was built using the CGIDEV2 tools. 
It's been running almost faultlessly for nearly 2 years using the 
"classic" IBM HTTP server. A month ago I migrated the HTTP server to 
Apache and again it seemed to go very smoothly.

At night the sight is taken down for 10 minutes to run a back-up and 
refresh, the dayend job issues an ENDTCPSVR, waits for 60 seconds then 
issues a STRTCPSVR to begin another APACHE server instance  that serves a 
static "Sorry, site down for 10 minutes...." page. Both of these server 
instances share the same port but both would normally never be up at the 
same time. This has worked fine all of the time. 

However, occasionally one of the QZSRCGI jobs doesn't end and just sits 
there locking files (all of the jobs hold locks all day as the files all 
remain open as soon as the  server starts). The job isn't looping, it just 
sits with "TIMW" status and occasionally goes to "RUN" status. The static 
page instance works fine and successfully serves the closed site page. 
However, of course, the dayend crashes because the rogue server instance 
is locking files that need to be backed up. Subsequent text messages to my 
cellphone wake me (and our baby) up at 4am to tell me that it's crashed, I 
usually log in and put a 4/*immed against the job. It usually takes about 
5-10 minutes to end which is very odd because the entire server usually 
ends almost instantly when it works correctly.

All of the PTF's are up to date as I did this at the same time as 
upgrading to Apache.  I've looked at the job logs and they look exactly 
the same as ones for a normal instance. There are no related QSYSOPR 
messages either. I also checked history logs and usually one or two of the 
other QZSRCGI instances took 10-20 minutes to end although the other 20 or 
so ended okay.

Has anybody encountered this or have any ideas on how I could try to 
resolve it as I'm getting a bit weary?

Thanks

Mike Bailey