That is a cool trick. In real life of course XML::Simple (or any of the DOM modules) would probably be unsuitable for processing the log files since they all create a tree representing the whole contents of the file in memory. XML::Twig would be a fine choice or alternatively a SAX approach would work too.
On the other hand, as merlyn pointed out, YAML might be a better fit. While the extra 'fluff' of XML can compress well, you will have to uncompress the whole file to process it (ie: without root elements, you couldn't parse the XML from a unzip stream).