> The problem is that it downloads pages even if its
> content has not changed at all.

Your script should:
- produce an "Etag" HTTP header with some kind of content-identifier (this can
be the hexadecimal md5 of the page that is going to be flushed, or more
cleverly the hash of all dynamic and external parameters used to produce the
page)
- check before flushing the content that the client did send an
"if-none-match" header, and, if the provided etag is the same as the one you
just computed, do not send any content, but a standard HTTP status code 'not
modified' (304)