Seven workloads under Node 22, 24 and 26 on one machine, interleaved to cancel drift. Only two differences survived the noise: JSON.stringify is 25% faster, and Map insertion moved around rather than improving.
AsyncLocalStorage adds about 250 nanoseconds per request on Node 24 and 26, down from 510 on Node 22. Reading the store costs roughly 25 ns a call. Measured across three versions with the script included.
A worker pool is 35 times slower than the main thread when each task needs 0.05 ms of CPU, and 3.5 times faster when each needs 222 ms. Twelve measured cases show where the line sits and why transfer beats structured clone.
The same node:http server run under Node 24 and Bun 1.4, measured over three rounds. Bun serves JSON 81% faster and hashes 55% faster, and on a plain text response the two are the same.