dive into mark

$ lynx -head -dump http://hixie.ch/advocacy/xhtml | grep Content-Type
Content-Type: text/plain; charset=utf-8

According to comments by Jens Alfke, the next version of Safari will extend this behavior to sniff for feeds as well:

Also, there is a bit of code way down in WebCore that sniffs the incoming page and, when it detects the start of an XML document that contains RSS or Atom, it auto-corrects the MIME type to application/xml+rss or application/xml+atom.

Words fail me.