As mentioned on llvmdev, the 'firm' project recently released the source (GPL2 license) for their new C parser, named "cparser":
In my brief look at it, it looks even more half baked than Clang :), and (like firm) seems to have an "academic" approach rather than being built with a goal of a fast and fully conformant frontend. It currently doesn't have a preprocessor (though one is in progress), doesn't track column numbers, uses global variables all over the place, does not have an optimized lexer (or anything else), etc. OTOH, it *does* have a man page, and can apparently bootstrap itself!
I'm not sure why they didn't just use clang, but being written in C++ and/or using a BSD license seems like possible reasons.