Yes, all mathematical niceties don't matter if MMSE estimation is not a good estimation in practice. This is the reason Gauss invented MMSE in the first place along with the eponymous distribution, while working with astronomical observations iirc.