High number of CONNECT_FAIL errors

Recently our squid cache (previously 1.NOVM.20+retry, also tried 1.NOVM.21)
has started returning an extremely high number of ERR_CONNECT_FAIL's.

This started approximately 48 hours ago, and so far as I can tell, doesn't
coincide with anything else that was changed on the server (Solaris 2.5.1 x86).
Prior to this, the cache has been running fine in every respect. Setting
browsers to not use this cache works fine.

Any ideas where I should begin to start looking? Memory and file descriptor
usage appear ok...

-- 
Phil Dye                        | Work: pmd@tcp.net.uk
Network Manager                 | Play: phil@lart.ing.co.uk
Total Connectivity Providers    | Consider myself properly disclaimed 
"You can lead a boss to a decision, but you can't make him think" - NWW