We consider a model in which the ultra-relativistic jet in a gamma-ray burst (GRB) is cold and magnetically accelerated. We assume that the energy flux in the outflowing material is partially thermalized via internal shocks or a reverse shock, and we estimate the maximum amount of radiation that could be produced in such magnetized shocks. We compare this estimate with the available observational data on prompt gamma-ray emission in GRBs. We find that, even with highly optimistic assumptions, the magnetized jet model is radiatively too inefficient to be consistent with observations. One way out is to assume that much of the magnetic energy in the post-shock, or even pre-shock, jet material is converted to particle thermal energy by some unspecified process, and then radiated. This can increase the radiative efficiency sufficiently to fit observations. Alternatively, jet acceleration may be driven by thermal pressure rather than magnetic fields. In this case, which corresponds to the traditional fireball model, sufficient prompt GRB emission could be produced either from shocks at a large radius or from the jet photosphere closer to the center.