On the validity of the mean field approach for the description of pairing collapse in finite nuclei