I do not monitor as much as you do, roughly 485 devices. I am only interested monitoring WAN and LAN links. Not too fussed about pc`s and printers. Due to internal politics, I steer away from servers too.
I use a single poller too, NPM, NCM, Netlow< IPSLA and APM (just for the SW box
64bit 2003 with 1TB disk space, 30Gb Ram
it all depends on what you can give up detail on.
if you can accept polling volumes once per hour that will relieve some of stress
but the fastest and quickest way you can affect this is to increase your polling interval on interfaces since you have 13245 of them
interfaces that go to users are usually a good place to start, that way you can keep the faster polling for the wan interfaces
the effect this has is that you will have flatter graphs on the utilization losing some detail in where data spiking might be occuring.
slowing your statistics on interfaces to 14 min will give you some breathing room and then you can think about slowing the status polling on less important interfaces to 4 min intervals.
Like i said in the first line it all depends on what you can give up in details.