Tell that to wikimedia, I've used libxml's SAX parser in the past to parse 80GB+ xml dumps.