The primary difference between these two functionals is that wB97XD can calculate dispersion forces like hydrogen bonds and dipole dipole interactions but B3LYP cannot.
I don't know which one is good for metal oxide between these but on safe side I would choose wB97XD. Also, my labmets performed benchmarking calculations on gold nanoclusters with m062x, PBE, MN15 against CCSD/LanL2DZ. They found that MN15 provides best results. (Using gaussian 16)
You should also perform benchmarking calculations against CCSD(if you can afford) or against MP2 with basis set you are planning to use.