But that's not what they were measuring. They measured sales revenue per employee and then admitted in errata that the claim was misleading anyway since other factors also contributed to sales revenue increase.
In software it is much easier to scale sales: the infrastructure costs when selling 1M copies and 1.4M copies are quite similar. Maybe some devops will have to spin more servers.
But programmers allready did their job.
Compare this to selling physical objects. If you can make 1M cars per month, you cannot easily make 1.4M per month: it requires more workers, machines, supply. Those are quite neglible in software.
Also your workers can make cars 2x faster than everyone else (better productivity) but nobody will buy them, since they are poor quality
