As an aside ... one of the vexing problems that I often have with XPath ... powerful though this facility is ... is to get the dammed query exactly-right.   Web sites such as http://www.xpathtester.com/xpath can be priceless, because they let you stuff some data in, and (attempt to) run XPath queries against it, and see what happens (or, what errors are returned).

That way, you’re only banging your head against one brick-wall at a time:   first, getting the XPath query correct and knowing what results it should be producing; then, getting it to run in your Perl and to produce those same results.   (And it’s a reasonable guess that libxml is being used in both cases.)