Real-world testing of squid-2.5 NTLM

On Tuesday Robert and I found and fixed a cache-corrupting bug which caused
my stress-test (2.5DEVEL-ntlm) server to die after an average of 250k hits.
Now my test server has been up and runnning for 1.7M hits, and it's faring
fine.

My main concern now is with memory consumption: squid appears not to be
leaking memory,
at least not in a noticeable way. However, memory usage is EXTREMELY high.
Just to make up some figures, "top" reports that the squid process is now
236 Mb big,
and this is the relevant configuration information:

cache_mem 5 MB
cache_swap_low 90
cache_swap_high 95
cache_dir null /test/cache

All other parameters use their default setting.

Attached is the HTML output of the cache manager MemPools info.
Notice the unordinately high number (and thus mem usage) of idle
Store Mem Buffers.
That value alone makes up for 95% of the MemPool-used RAM, and for 60%
of the total RAM allocated by the process.

Ideas?

-- 
	/kinkie