The problem isn't runtime, it's the memory. The earth is the most efficient storage of all the information about the positions/trajectories of all particles that make up earth.
That's missing the point, I think. In the simulation the Earth doesn't exist. Instead, some high-level model is used to generate the observations we make at the macro scale. Only when we perform sensitive experiments which require a finer resolution is that detail generated in such a way as to remain consistent with observations made so far.
Of course as someone who writes simulations I can tell you that it is not so easy in our computational models. Indeed it would seem very difficult, though not fundamentally impossible for a high level model to correct macro behavior in all cases. But we cannot know the constraints of the universe which contains our simulation, which may be very different from ours.
