But then wouldn't this require a bigger team with vast expertise or a lot of time compared to a more normal approach, and it will only result in marginal performance improvement, which might not be important in most of the regular projects?
"Best language for a task" isn't meant to refer to performance of code (except secondarily when the task is highly performance sensitive) but rather the language in which it is easiest to accomplish the task.
