Want proof? Recently, I spent just a few minutes crafting an HTTP server using Boost.Beast. Each request can send Lua code, and for every request, a new Lua VM is instantiated, the code is executed, and the result returned.

Think of it like AWS Lambda — but without Firecracker’s isolation.

I ran benchmarks with over 10,000 parallel requests — that is, 10,000 isolated Lua VM instances orchestrated on just 8 physical cores, with a latency of around 15ms.

And that was using code written by ChatGPT. If I dove into aggressive optimizations and coroutines, I could probably reach 100K simultaneous clients — assuming the kernel allows it.