Only in complex routing or services provision situations have I seen a NEED for periodic "cleansing". This does not hold true in an unstable environment, such as during or immediately after an attack, during major restructures, addition of services, etc. Then its a good idea to flush the garbage from your routing, CAM and ARP tables.
During annual Disaster Recovery Test, as well as while rolling out new IOS implementations we do power cycle our core (half at the time) so as to verify failover and operability with the new IOS.
The only service utilization on a router that I've seen that may gain from a periodic reset is those assets associated with PSTN call switching. We've been able to clear multi-interface call trunks, dial-peers and the tables associated with the service faster with a reload that can be accomplished via the standard administrative commands. We reload these routers (3660s) about once every 2 months as a rule of thumb.
The reason here is that individual DSPs can "hang" the LEC D channels may be mis-provisioned, and this clears the DSPs and can help identify issues with the carrier (we all know they never change anything that can cause a problem).
As the other posts have stated, Cisco gear can run for a year or more with no problems. BUT if you are seeing via "show proc mem" or other means that your memory utilization is creeping up, you may want to institute a scheduled reload vs seeing a unscheduled, mid-day, while the CEO trying to get e-mail and the CIO is in video conference, MALLOC Error.
