HTML-TreeBuilder-LibXML is a life-saver: finally something that combines the HTML-TreeBuilder interface and XML-LibXML, and gives one XPath in a sane way and does not use lame stuff such as XML::XPath.
Now all we need is something like jQuery or HTML::Zoom and we will reach the Perl XML/HTML parsing nirvana. You may also be interested in my WWW::Mechanize::LibXML , but that's kinda a hack, and I feel that WWW::Mechanize should make a more integrated use of LibXML, XPath, HTML-TreeBuilder, etc. under the hood in the built-in distribution. (If I only had the spare cycles for everything I wanted to work on, but naturally feel free to offer a small bounty.)