"The model uses physics at the one-millimeter rock grain scale to explain how the whole planet behaves."
A 3,000 x 3,000 x 3,000 grid is considered very large for modern scientific models. Assuming they are working on a cartesian grid, and an earth diameter of 12,000 km, their model would be 12,000,000 x 12,000,000 x 12,000,000; twelve orders of magnitude larger than the biggest physical model I've ever heard of.
This cannot be the case.
Whew! Its a good thing they never claimed they were doing any such thing.
"physics at the one-millimeter rock grain scale" does not mean that they were using a model grid of that same scale.
To show that the assumption that it must, or even should, is incorrect consider any engineering model that involves the effects of static friction.
The phenomena that cause static friction exist on the molecular and atomic level, and theoretical predictions of friction under arbitrary conditions need to be analyzed and calculated at that scale.
But once you have determined what the coefficient is under a given set of conditions, you only need to use the single number in a macro scale model.
That is what they did here. If you read TFA you will see that their macro model used coefficients calculated using the detailed physics.