More squid 2.0patch2 filling past the limit

Running Squid 2.0patch2 has been pretty good thus far, but I am stumped by
one problem.

Since I switched to using ASYNC disk I/O, squid's not deleting items as I
think it should. Obviously, unlinkd is no longer running, but the amount
of used diskspace never shrinks.

Right now, despite my config directive of :

cache_dir /opt/squid/cache 2500 64 256

Squid's already used 4.5gb. I'm a little stumped, especially as it's
running fine on another machine of an almost identical configuration. The
only difference between the two machines is the load level, which is quite
high on the one that does not delete...not as much on the one that runs
fine.

Could this be because all 32 threads for AIO are in use at all times?
That doesn't seem likely, as things are not reporting such issues in
cache.log.

Is there data I can interpret from my store.log? Is there a way I can
find out what the threads are doing? Other than cache_mgr.cgi?

I'm running on HPUX v11.0.

-- 
mat.