XML pages are usually going to be well formed, and less likely to change than HTML as you won't have look & feel noise getting in the way of your scraping; however the strategy discussed for GjScrape still may apply.
Failing that, use an XML parser of your choice.
It's non-standard, but I like to use gj-xml; a small, fast and easy API: http://www.osjava.org/genjava/multiproject/gj-xml/.