My assumption here is that the Node implementation loads your code off disk during cold start using standard require(). Whereas the Go implementation requires loading a small wrapper (probably also in Node) which execs the Go process which then handles your request.

In my experience, GCP Functions Go cold start time is almost nil because they compile your code into the function entry point directly, so it's as quick as calling a function in process.

Forgive me if I'm wrong, but I believe it's because Node is implemented with V8 which is a JIT compiler rather than just an interpreter, not to mention that Node is a more mature and optimized runtime than the Go toolchain.