The nagios system is modular, and based on cron jobs, and a few daemon processes (from what I remember). Since it is modular, you only need to activate the monitors for the services you wish to maintain. Load is very minimal. Most of the systems I ran it on were PC on a card type systems using a Cyrix Geode 300mhz CPU (Spec benchmark equivilant to a Pentium 75), with 64M memory. The application we were running used 50M, and sometimes an engineer would remote in with vnc (using KDE or Gnome as their windows manager). Those were the times when I got alerted to excessive cpu load & swap load. At one point it was so bad (one brain dead engineer in particular), that I had to rebuild the system image with only mwm as the windows manager (talk about pain and suffering). Most of Nagios runs from a single monitoring system anyways. There were some plugins that I wrote in perl to monitor system throughput of our test automation system, but they became redundant as we integrated that capability into our test software.
Other than that, I had ~140 systems running automated testing & data upload, 2 servers running Apache & MySQL (mirroring each other), and a handful of xwindows servers for remote login & development (not to mention a few "ghosts" on the network, like an MP3 streaming server).
Nagios actually performed well back then, but it had fewer components than it has now. It looks like they have made some major enhancements since then (2001 was when I last implemented it, before my job became redundant).