I happen to think that load time is important, and the time to fork should impact all tests similarly.BTW XML::Twig does probably very badly in this respect, so you can't say I am biased.

As far as I know, no one has ever challenged the "SAX is lightweight and fast" before I published this benchmark. And no one since then has ever come up with any figure that would prove me wrong when I say "SAX is slow".

Of course my benchmarks are imperfect. Of course I am sure you could do better. Then do it.