Hi, Max! I can see that working. I'll see if the overhead incurred with the extra layer is less than the cost of calling the fastcgi over and over. My guess is yes. Some question come to mind for the extra layer:

1) Is it better / faster to use a socket for the proxy_pass to itself? Can nginx be set up to listen to a unix socket?

2) Would I gain much benefit by using an upstream keepalive module for the managing the connections to the same server? Would your ngx_http_upstream_keepalive add-on do the trick?

Perphaps in the long term, a clone to error_pages that caches along the way would make sense. Maybe an add-on that I'll take up when I'm less busy.

Thanks for the reply. Nginx is a great piece of work!