Allow me to toot my own horn here a bit: I have developed a SAX2 parser in Java that is designed for on-the-fly parsing of tag soup; it's called TagSoup, naturally (http://www.ccil.org/~cowan/XML/tagsoup).
As delivered, its tables are designed for ugly HTML, but it wouldn't be hard to create tables for sucky RSS as well. Anyone want to cooperate with me on this?