>As you point out, there is a difference between the median performer

But I didn't say that, and I don't agree.

I think the 90th percentile never causes delays on a particular station, while the median may cause a few per hour, and the 10th percentile might delay every 2nd or 3rd cycle.

There are tangible costs associated with those levels of performance.

This is a completely different argument though and irrelevant to the comment chain you replied to. Everyone involved may even agree with you - why did you frame it as an argument against the claim that there is less capacity for 10x performers?

Generally you design your cycle times so that the median performer never is late. Remember if each median performer has 1 delay per hour, on a line with 100 workers that mean each cycle there is a delay somewhere and so the line always moves slower than expected. How much slack to add is a complex question, but you have to assume there will be delays. The 10th percentile cannot be allowed to delay the line, so you have to deal with it (training, fire them, give them faster tools, or switch to a station they can do in time come to mind, there may be more)