Where I work now we have a scheduled monthly maintenance window. On a particular Saturday we get to do whatever we need to the boxes. Everybody that uses the systems knows this and nobody works that Saturday.

Our policy is any system which hasn't been rebooted for 90 days is rebooted. Oracle servers and some Veritas Cluster Server clusters are rebooted each month instead. There are a few production boxes exempted from the policy because they support manufacturing plants which have shifts working on Saturdays.

As for the reason why - it is basically just for cleanup purposes. Stale NFS handles, zombie processes, small memory leaks from applications, etc. are all cleaned up by the reboot. Also it is a way of testing to make sure everything runs smoothly with the startup and shutdown scripts so we don't find problems with them when there is an unscheduled reboot.

I've worked other places that never rebooted machines unless it was for a specific purpose - hardware upgrades/repairs, software installs which require it, etc. It all depends on the environment and what the system admins prefer whether a scheduled reboot is appropriate or not.