Thanks for the HTML4 standard link; I wasn't aware it didn't allow special characters in URLs. I have modified my static site generator to percent-encode these characters; it may be unimportant but I may as well conform to HTML4.
In HTML by the WHATWG they added exceptions for UTF-8 URLs, but I don't think is a good idea. This breaks software that doesn't handle UTF-8 URLs (i.e. anything that follows the RFC not what Google says).
Can you name any examples of software that would break when presented with un-encoded UTF-8 URLs? Even Lynx supports them. P.S.: I know about RFC 3986; I've implemented it :^). The source of my confusion was that HTTP 1.0 (RFC 1945) defines URIs in a way that treats bytes >=0x80 as unreserved (section 3.2.1), and I falsely assumed HTML must be the same.