I want to self-nominate html5gum. It's a HTML5 tokenizer compliant with the WHATWG HTML spec, meaning that it can parse most real-world, malformed HTML like browsers do. It's stil work-in-progress because I'd like to expand it to become a full parser that understands tag hierarchy, but in its current state it can already be used to extract attributes from tags, and things like that.
It can be used to parse HTML significantly faster than html5ever's tokenizer (when using lower-level emitter API, otherwise it's slower atm), and only 2x slower than using quick-xml's SAX parser.
I'm currently working on getting html5ever up to speed with recent changes to the HTML spec such that comparisons can be done more fairly.