We study the three-spin model and the Ising spin glass in a field using the Migdal-Kadanoff approximation. The flows of the couplings and fields indicate no phase transition, but they show even for the three-spin model a slow crossover to the asymptotic high-temperature behavior for large values of the coupling. We have also evaluated a quantity that is a measure of the degree of non-self-averaging, and we found that it can become large for certain ranges of the parameters and the system sizes. For a spin glass in a field the maximum of non-self-averaging follows a line for given system size that resembles the de Almeida-Thouless line. We conclude that non-self-averaging found in Monte Carlo simulations cannot be taken as evidence for the existence of a low-temperature phase with replica symmetry breaking. Models similar to the three-spin model have been extensively discussed in order to provide a description of structural glasses. Their theory at mean-field level resembles the mode-coupling theory of real glasses. At that level the approach via one-step replica symmetry breaking predicts two transitions, the first transition being dynamic and the second thermodynamic. Our results suggest that in real finite-dimensional glasses there will be no genuine transitions at all, but that some features of mean-field theory could still provide some useful insights.