Yes and no; while it's still true that you can write FORTRAN in any language, there are network effects that mean the effort required to write efficient code is different per language/community/framework.

I think that’s true if you take a no dependencies approach but as soon as you tap into the ecosystem the argument is lost.

Not really. It still depends on who coded it. It's just that there's more people involved now.

Corollary: The average API is far less efficient than the languages you like to complain about their efficiency.