I thought it was from Peopleware. They claimed they found the result by performing coding wargames among professional programmers in the 80s.

They claim that the best are 10x better than the worst but only 2.5x better than the median. Additionally programmers from the same organizations had very similar performance. I.e. there probably isn't a 10x difference in performance within your company.

The book claims similar distributions when you measure many kinds of human performance, so I wouldn't be surprised if there were many studies that replicated the general idea.