The article is quite good but the Node.js example is wrong. OP is measuring dead code elimination and general overhead time.

No strings get harmed in the process, run Node.js with --trace-opt to see what's happening.