Node is best-suited for when your API itself does very little CPU work. When it serves mainly as a task-delegator to a DB, other APIs, local disk I/O, etc. All of these are non-blocked by JS and automatically handled by underlying native code in parallel. The main advantage of Node compared to Go is that your front- and back-end can be in the same language, which helps with hiring and can even allow for code-sharing.

That said I don't think Node would ever match Go on raw performance, even in its wheelhouse. It's all about tradeoffs.