If that is your thesis,

It isn't. Just a guess that fits the observed data.

you picked a really lousy experiment to test it.

Had that been my purpose, I would have agreed with you.

Luckily, it is easy to disprove.

No point. You're disproving the wrong thing.

The test was simply to assess what reduction in time availed by -u when sorting a file with a high proportion of dups.

What it proved is: negligable!