This shouldn't be too expensive, since the connections are pooled,
but it should save us some bandwidth since we won't fetch non-html
files and files that are too large for us to process (especially
since you can't cancel a request without closing the connection
with HTTP1).
