Well, it all depends on your application. If my script were actually an IRC bot, I'd go with a simple regex-based approach as described above. In that application it's important not to download 100K at all (risk of DoS-attacks), even if the header is that long.

I know it's evil to parse HTML with regexes, but sometimes it's simple and convenient, specially if you are happy with a solution that works in 95% to 99% of all cases. Note that all markup, even comments, are disallowed in <title>...</title> tags, which simplifies the matter.

Of course things are different for more serious matters - if you want an application that extracts the title of all valid HTML pages (and most invalid ones as well) with an accuracy matching that of the w3 markup validator you'll have to download it all.