Note that, html5lib handles self closing a, p, div, table, body and even html tags correctly. Handling <title/> as an instruction to treat the rest of the document as text is the wrong decision. It is *never* going to be right thing to do. But, given that browsers make this same mistake, I'm guessing you are not going to change this. Oh well, a spot of regex to the rescue.
html5lib is behaving correctly per the spec. of the HTML parsing algorithm. When the tree builder sees a start tag whose tag name is "title" while in the relevant insertion modes, it puts the tokenizer into the RCDATA state.
Sure, I can believe that, given that browsers do the same thing. If it were up to me, I would depart from the spec on something that is so obviously wrong. But, I can see the value in having a parser that hews closely to the spec. I can always use a regex/patch html5lib myself for my use cases.