Do multimodel ensembles deliver a superior posterior in paleoclimate data assimilation?