The results obtained depend critically on the quality and size of the basis set, where more is better. This is however not true for spin-state energies as shown by us recently in J. Phys. Chem. A 2008, 112, 6384-6391. There we showed that STOs work fine, already with small basis sets, i.e. they converge fast. Large GTOs give the same result, but the convergence is much slower. And basis sets containing effective core potentials (ECPs) give systematically different (wrong!!) answers, so should not be used for spin-state energies.
Small Pople-type GTO basis sets (like 6-31G* etc.) are in particular very poor. We have studied in this paper (J. Phys. Chem. A 2010, 114, 7191-7197) these basis sets, and showed that their faulty behavior can be largely corrected by increasing the [2d] contraction scheme to a [3d] contraction. In particular what is needed is an additional diffuse d-function. The resulting spin-state-corrected GTO (J. Phys. Chem. A 2010, 114, 7191-7197) basis sets give good results, and are available from the Basis Set Exchange online database.