The scaling behaviour is actually surprisingly good. The time to simulate quantum mechanical systems generally grows exponentially with the number of states, i.e. the number of atoms in this case, but the listed running times grow only somewhere between linearly and quadratically with the number of atoms.