A recent article offered that servers should be rebooted on a regular basis and mostly argued the case of OS updates requiring this. I think that servers should be rebooted/shutdown but only in the region of months or bi-annually. 3 reasons:
a. applying of kernel patches ( unless you’re using ksplice on Linux )
b. disk failure
c. memory fragmentation
In the first case, kernels can only be patched ( with the above caveat ) through a reboot – thankfully serious kernel issues ( in the Linux Kernel ) are rare. Other OS’s require restarts more often even though kernel updates may not have been applied.
The 2nd case deals with the reluctance of some disk drives to start rotating once they’ve been running for long periods, have worn grooves in the bearings and have cooled down after a shutdown. The drives effectively seize on restart. So it’s a good idea to physically switch off power to a machine ( rather than just reboot ) at intervals and let it cool down completely.
The last option deals with the natural fragmentation of data structures in memory after periods of time. Most modern OS’s though have mechanisms to deal with this so reboots here is not essential.
To advocate reboots just for the sake of it seems to me a little bit of preventative maintenance for those OS’s which are less stable and require this sort of thing. Otherwise, once or twice a year should suit you fine.