When I load this site, the text seems garbled. When I load it in FF, it starts garbled, but then switches over the page to the correct character set.
This document has a broken structure (part of body content comes before head). In such cases, WebKit only checks the first kilobyte for meta charset - while Firefox will restart decoding if it sees the meta at any place in the document. HTML5 specifies Firefox behavior, but despite this, I have very strong doubts about it, because parsing can have side effects (like script execution), which re-parsing would repeat.