[array_key_first]

You don't need proof or measurement if it's a question of time complexity. The time complexity covers that.

And if you aren't optimizing for time complexity, then it's probably an over-optimization. Even if switch was faster, how much are we gaining here? A microsecond?

But, I've had instances where I've turned reports from minutes to milliseconds because of some shit SQL or doing something crazy in a loop. That's the optimization that matters, IMO.