Install and configure Squid as a surrogate on the same system where you run an HTTP server.
Make a few test requests with squidclient. Pay particular attention to the reply headers and notice how the requests appear in both access logs.
Try to poison your own surrogate with fake HTTP requests. It is probably easier with httpd_accel_single_host enabled.
Estimate the size of your origin server's document set. What percentage of the data can fit into 1 GB of memory or disk space?