This paper compares different methods for removing resonance effects from second-order perturbation theory calculations of vibrational energies in a variety of systems containing from two to six modes. Both the recently proposed method of Kuhler et al. and the standard approach of Nielsen yield stable energy levels even very close to resonance, with the latter giving smaller average errors in such cases. In addition, the method of Kuhler et al. is observed to affect the ground-state energy, unlike the standard approach. This generally worsens the accuracy of the vibrational partition function at room temperature, especially for systems close to resonance.