I wish it were a joke, but we recently had a refund request because of high loads only to find that the user was running the unixbench test during the high load periods. (His normal load average was close to 0)
He complained that the load average was over 10 and was setting off his alerts and he contributed this to poor performance and wanted a refund. After trying to explain to him that unixbench runs multiple threads and would cause the load average to increase greatly and failing to get through to him, I have granted the refund.
Thought I would share my story and hopefully people realize that load average has nothing to do with performance if you are running a benchmark program that is designed to utilize all available CPU.