This gives better insight in the performance of the evaluator itself.
This includes two benchmarks: one for parsing XML and one for evaluating XPath.