The main purpose of sets is FAST membership testing. Everything I've seen in prior discussions convinced me that preserving insertion order would slow that down.
If membership testing lost even 10% speed, I would be -1000 on the idea. If someone comes up with actual working code that is no more than 1% slower, is move to -0.
If speed miraculously got faster in such a rewrite (analogous with dictionaries whose order was initially a happy side-effect, but I've seen explanations of why that wouldn't happen) I'd be +1.
But in the last case, the entirely of my support would be for the speed up, and exactly none of it would be for the ordering.