But their very point was that the simulation wouldn't have to compute every single individual particle, only the "interesting" ones. The rest would be approximated to whatever level required.

I basically agree with you but I can't see how your argument applies here.