I don't believe it's legal to have an HTTP response without a valid Content-Type. A text/plain content-type, though, is fairly common as a "fall-back" for most web servers.
The problem here is that IE never trusts the web server. When it receives what it considers a "possibly too generic" content type, it will second-guess the web server and try to figure out the content type on its own, so if it sees a text/plain content-type, but something that looks XML or HTML-ish, it will interpret it as XML or HTML.
Most people view this as really evil/stupid behavior, since it encourages misconfigured web servers. Neither Mozilla nor Netscape have adopted this practice, to my knowledge.