Note that this adds some significant overhead.

Have you measured it? I have. It is far from "significant" unless you have at least one hash that holds a very large number of keys. I find such to be pretty rare in the JSON I've had to deal with.

I have found much more overhead resulted from the pain of having to deal with random order of keys in JSON, making comparisons or just visual searching much more difficult. And overhead to human processing costs a ton more than overhead to computer processing, IME.