The request itself is always ASCII - that's the spec.

However the URL-Encoding scheme with %DE%AD%BE%EF encodes only bytes, so you need to pick a character encoding.

When I used latin-1 for this encoding some browsers sent me some requests with latin-1 encoded data, even though the pages themselves were encoded in UTF-8 (and declared as such). So I guess they decoded the URLs and did some encoding guesswork, and used that for GET requests.

Which is why I recommend consistency ;-)