First would type out a few lines of code incredibly fast with lots of mistakes (and not bothering to backspace over wrongness) - then go back and clear up damage using notepad level editing.
Second would stare fixedly at screen for several seconds, whilst remaining still as a statue, then carefully emit the minimum string of editing commands to achieve the change.
Their actual rate of progress was pretty much identical.
That eliminates many of the benefits. Was your coworker relatively new to Vim? Or maybe pondering what change to make, not how to actually make it? This is what I've seen from a coworker who was learning Vim and forcing themselves to learn the "proper" way to do things. I don't consider that to be a bad thing, provided that you balance it with actually getting work done.
I always do whatever allows me to edit the line without or with relatively short pause. That might be efficient vim commands, or it might be sloppy if the line isn't formatted in such a way that it fits my muscle memory. If I'm dissatisfied with the edit and think there might be a better way, I research it after the fact so I can later practice and know for next time.
This comes with experience; I've been using Vim for over a decade. Any pause in my editing is because I'm actually figuring out what change to make, not how to make that change.
As you rightly point out - there is a reasonable middle ground.
There's definitely an upper limit to human text manipulation speed, and I would be downright shocked if the difference between "fastest vi user" was even a single order of magnitude better than "experienced mouse user."