I think this might be related to the pareto distribution of productivity in my line of work (steel fabrication). Our most productive employees are literally 10 times more productive than the least productive. And it's always the same people at the top of the list, and always the same people at the bottom. If you pair them together, the productive employee loses 10-20% of productivity while they're together, and the less productive employee improves by 5-10%, but as soon as you separate them, they go back to where they were before.

I'm curious about this.

One question is what happens if you pair two of the top performers or two of the bottom ones? My prediction is that nothing changes.

The second one is related to the point of my comment. Can you build a graph date/performance? That could show curves similar to the ones of the go players and show how long it takes to flatten the curve.

If you pair two top performers, they get slowed down by 10-20% because they each have their own way of approaching the work, and there's usually conflict as they are each convinced of their own way. They work even faster, however, if you put them next to each other, but working separately. They watch each other and start tacitly competing. You'll get another 10% out of them generally.

As for pairing the two lowest producers, they'll sink even more as they are confused by what the other person is doing and it slows them down.

Changing someone's level of productivity is essentially untrainable in my experience. It's not a training issue. The people at the bottom know how to do the work, but any attempt to show them a more efficient way of doing it is overwhelming and confusing to them, and they shut down.