VCL reference in the worker thread when the request is done and
with a cheap check reuse it for the next request handled by this
thread.

This should reduce mutex contention.


git-svn-id: http://www.varnish-cache.org/svn/trunk@1032 d4fa192b-c00b-0410-8231-f00ffab90ce4