Selective parenting

Situation: We run a squid proxy locally, with the parent being our uplink's
filtering proxy. This parent proxy (Bess) is *extremely slow* (probably
overloaded). Here's my solution: make the assumption that if
"http://www.foo.com/x/y" is blocked (or not blocked),
"http://www.foo.com/x/.*" is also blocked (or not). This proxy adds the
header "Pragma: BESSBLOCK" to blocked sites (the content is a fairly standard
"you can't go there" page.) I don't have real access to the blocklist.
Getting only headers through the other proxy could speed this up a bit more
(all accesses would be direct), but could be a bit more complicated.

Based on that assumption, I should be able to keep track of unblocked sites;
if a URL falls under a previously unblocked path, bypass the parent. Only
the first access to a given site will be slow. (Remembering blocked sites
isn't needed.)

Is there any way to do this within Squid's current architecture, or will I
have to break out Squid's source? Would it be easier to write another
simple proxy to do this (selectively using Squid or the filtering proxy
based on the same conditions)? Has this (wishful thinking) been done before?

-- 
Glenn Maynard