Regarding letloop http server, I added CML, and fixed a couple of bugs, along the benchmark now do apple-to-apple comparison (template vs. string interpolation, one CPU core), here is the summary:
On the counter-page scaling benchmark (wrk, 1–256 connections, each server single-threaded and pinned to one core, performance governor), the peak throughput each stack reached — from slowest to fastest — was: FastAPI/uvicorn+uvloop at 60,873 req/s (64 connections), Node.js at 189,890 req/s (8 connections), Go net/http at 224,850 req/s (8 connections), bun-rust at 320,226 req/s (32 connections), Rust axum/tokio at 404,789 req/s (8 connections), and letloop (amalgamated whole-program build) at 452,542 req/s (32 connections).
Averaged across all nine concurrency levels, letloop's speedup over the field, from its fastest rival to its slowest, is 1.10× over Rust, 1.37× over bun-rust, 1.92× over Go, 2.25× over Node.js, and 6.74× over FastAPI.