How can you effectively use the idle time of processor cycles across a network

These projects typically operate by installing an agent on a node (usually a regular desktop computer) that registers over a local network or the Internet with a master server. The master server is usually running a program designed to divide the computational work into small chunks. The server will then send small units of work to available participating nodes. The agent on the participating nodes will monitor CPU resources and launch one or more processes when CPU cycles are available to receive the work unit, process it, and return the results to the master server. The agent will usually monitor its own resource usage and reduce its priority in favor of the local machine's primary workload (whether it be a network server or a local desktop machine running word processing software).