My DOM parser tests compared XML::Simple vs. XML::LibXML in two platforms Solaris 10 and Linux 2.6, all my tests in Solaris ( different documents with different sizes ) showed XML::LibXML was slower than XML::Simple ( It was much slower accessing nodes), as for Linux it was the opposite and XML::LibXML was 8-10x faster than XML::Simple.

On the other hand comparing a SAX parser using XML::Parser vs. a DOM parser using XML::LibXML or XML::Simple, expat won in both Solaris and Linux.

Honestly, I haven't tried XML::LibXML::SAX and maybe it is way faster than XML::Parser.

This link its a bit outdated but It has some benchmarks. Probably things changed notably since then.

It was much slower accessing nodes

You seem to have missed "at creating the same Perl-land data structure." Visiting every node was included in the benchmark.

A task using SOAP::Lite that took two minute was sped up to 15 seconds when I replaced Expat with XML::LibXML as its parser. I kept the change mininal: XML::LibXML was used to create the same data structure that Expat was used to create.

Honestly, I haven't tried XML::LibXML::SAX and maybe it is way faster than XML::Parser.

As I mentioned, XML::Parser is by far the fastest existing backend for XML::Simple. I'm pretty sure I've posted benchmarks on PerlMonks.