Of course they are rolling randomly. However, they are rolling randomly, with the odds saying they are rolling some numbers a disproportionate number of times more than others.
Because the physical dimensions are not EVENLY inconsistent...that's not what "inconsistent" means. They are UNEVENLY inconsistent (which is a redundant phrase but apparently in this case necessary for clarity). So it will bias towards a section of the die, and away from another section of the die, due to the inconsistent shape of the die.
We know this because of the process used to create all polished dice. If the dice are coming out with different physical dimensions from each other (though they all come from the same casting molds so start the same size as each other), that is (always) caused by random polishing elements. Which means variable amounts of time in the polisher, and/or being struck by the rough objects in the polishing machine a disproportionate number of times and/or in a disproportionate number of ways. The odds this would create exactly even imperfections across exactly every single face and every single angle of the die are astronomically high. The overwhelming odds are the die is coming out uneven across the angles and faces of the die. And if it comes out uneven across those faces and angles, that means it's not rolling an evenly random roll when rolled.
I'm not denying there are imperfections in the die. Please stop explaining that. We get it, really.
What I'm saying is, those imperfections are not generally pronounced enough to make a statistical difference
in practice.
It's a lot like the difference between theorycrafting and actually playing D&D. The dice seem unbalanced on paper, but when you roll them, they perform just fine.
The only way to prove that a die is biased is to roll that die a bunch of times. Apparently someone here did that with a bunch of Games Workshop dice, so now I'm willing to believe GW dice are crap. He didn't say which style of GW dice he tested, though, so if it was the one with skulls for pips, I have to wonder why he wasted his time. You can tell by looking at the absurd design that they're not going to work right.
As for the guy who rolled a single d20 from each company 10,000 times, that also proves nothing. You need to roll a lot of different dice from each company to be able to generalize that one company makes good dice and the other does not.
But I will say this: No matter how carefully you shave off that sprue, unless you're using a laser, your Game Science die is
at least as cockeyed as the tiny, tiny imperfections found in the Chessex study, so there's some hypocrisy in action for ya.