I have some reservations about those parameters, though. If I interpret the table correctly, the nonzero charges for the "explicit gas diffusion simulation" are taken from heme oxygen. For free oxygen, this seems unphysical. The LJ parameters, on the other hand, are not the same as in toppar_all22_prot_heme.str ; they are from TIP3P , so they used charges from one irrelevant model with LJ parameters from another one...
For the "implicit ligand sampling", they used the correct heme oxygen LJ parameters and set the charges to 0. Yet, the original heme oxygen LJ were transfered from carbonyls in the protein force field, which might not work for free triplet oxygen. Contrary to commonly drawn lewis structures, the O-O bond is not a double bond, and the species has an unusual electronic structure.
Both models, especially the second one, feature O-O distances that seem too short compared to the reference value of 1.23 in toppar_all22_prot_heme.str and the experimental value of 1.207 for ground state (triplet!) oxygen.
Then again, I don't know of any well-validated parameters for free oxygen in solution. Generating them would be a more challenging endeavor than it seems. In the end, it may not matter that much; in practice, a nonpolar blob of approximately the right size would probably yield qualitatively correct results, though for quantitative free energy calculations, more validation would be required at the least.
Oh yeah, apart from the above reservations, I don't see any objection to using an existing atom type that has the same LJ parameters.