Based on discussion with Yutaka, in https://github.com/whatwg/mimesniff/issues/45 we're considering restricting MIME types to the same characters allowed by HTTP, but doing so in a way that we don't fail recognizing the following as HTML:

  text/html;
  text/html;?=x
  text/html;x=?

One consequence of this is that

  text/html;charset= gbk

would end up not recognizing the charset parameter as whitespace is not permitted in the HTTP token production.

I was wondering what you think of that bnc? (It goes a bit counter to what we discussed in https://bugs.chromium.org/p/chromium/issues/detail?id=772834.)

The advantage of this is that once we've parsed and then serialized a MIME type, we always know it to be HTTP compatible, including with non-browser systems.