Preloading libraries and caching is persistence.
From my perspective everything you say on programming matters is blatherskite. I'm not sure you even understand what CGI means at this point. You've definitely indicated a lack of understanding of FCGI more than once and even HTTP. But let's take you at your word one more time and give you a chance to defend the assertion you're making.
Cite facts to show your reasonable cost. Show how much the same code running in a pure CGI environment costs compared to running on any persistence server in common use. If you don't have them, here's how to
prove m –
My main app loads 1,325 modules at startup. Many more at runtime but let's be fair! So the test script should do something similar, perhaps just loading everything in Task::Kensho would be enough, as well as instantiating a DB handle, of course, and dumping the %INC or something in a template. Now, put it on an EC2 instance or something and report back on the cost difference, actual money and time both, running say 10K requests against each and let the audience decide if your conclusions are reasonable in cost; or any other way.