In a fully blown production environment this kind of thing is normally handled by centralised monitoring system (ala NetX). You can then write your own SNMP traps in whatever language you like on the client machines.

When a threshold or condition is met your agent would then fire off a trap to the monitoring server, which would do any number of things from paging, emailing, displaying an alert on a display or whatever.

Having said all that, it can be quite tricky to set something like that up as it involves defining business processes, standards and may be just plain overkill.

In that case, I like the simplicity of tachyon's approach using a cgi.