What was the size of the VM run queue during the tests? This can indicate if the server was overloaded in general or is the problem local to the repo pool.
You could also try working with the sojurn pool - it should handle working during overload better.
