rvt is pretty close, but for the us the high server load isn't caused by CPU overuse. From a theoretical perspective, server load represents how much processes are waiting on the run queue. Though usually the run queue gets really long because the CPU can't handle everything at once, our bottleneck is usually disk I/O. In short, our hard drives are too slow. Every request on our server requires that a file is opened up, requiring that process to stay in a "D" state where it waits on the hard drive(s).
At a typical moment in time, our hard drives are at around 90% utilization and processors are below 50%. Another interesting fact is that almost every one of our downtimes is caused by RAM running out and the system being forced to use swap (page file for Windows folks), which skyrockets disk I/O to the point where the server can't handle the load.