In my experience, there certainly exists a 10x gap between those who aren't very helpful and those who are extremely helpful. That is the crux of the research as I understand it anyway: You'll find a big distribution in talent, but a median producer is not 1/10 the best.

I've found those who produce significantly faster do so by creating well justified shortcuts that do not sacrifice the end goals (but may run around the requirements a little.)

That's exactly the story in TFA: Someone decided "I'm not going to satisfy this silly requirement of writing tests, I'm going to just auto-generate some tests, regardless of whether this has full coverage of all corner cases, because I bet everyone is just happier to move on."

This is why I believe a 10x-er can be trained, by building a deep toolbox of solutions to pull from, learning to read the room when it comes to designs, and learning to recognize and take well-justified shortcuts whenever possible.