What resource(s) are you running out of? CPU? Physical memory? Or is your main concern latency per request being too high?

Also, are you expecting usage to increase over time?

FastCGI should already be reducing CPU overhead and latency (similar to what mod_perl would do).

It sounds like you already have a platform that could scale across multiple computers; you would just need a front-end to load balance (a reverse proxy or a real load balancing appliance or even just multiple IPs and round-robin DNS in a pinch).