I recommend considering the XSLT approach, and tuen to using XML::LibXML and XML::LibXSLT, possibly with XML::SAX as the parsing API layer.

None of these will address the issue of having to work around inconsistent content and syntax. That's an albatross you're just stuck with. But the underlying parser libraries (libxml2 and libxslt) are quite fast and designed to work in harmony together.

--rjray