Would it help if our "nightly restart" for us in Australia was mirrored 12 hours later so there were 2 per day? If it's a matter of the server being online too long, perhaps another restart might help?
Over the years there have been some servers under my watch that even after the best of efforts put to those upstairs in the emerald tower were not replaced past their operating cycle but were very critical (a DC, Exchange and File Server AND SQL all in one, I inherited a nightmare) so restarting that beast was always scary because it had a bad habit of dropping the RAID during a power cycle, plus the programmers at one point had access to the server (again before I started) so there were... well it was a mess so sometimes Windows 2k3 would randomly say "No, just No... Not this time, reboot me again".
My point being I understand that restarts aren't always the solution and can cause more problems than you're trying to resolve in some cases. Just a thought