We investigate efficient and minimally unstable Pareto improvements over the deferred acceptance (DA) mechanism—a popular school choice mechanism that is stable but not efficient. We show that there is no Pareto improvement over the DA mechanism that is minimally unstable among efficient assignments when the stability comparison is based on counting the number of blocking pairs. Our main result characterizes the priority profiles for which there exists a Pareto improvement over the DA assignment that is minimally unstable among efficient assignments. We further consider an alternative natural stability comparison based on the set of blocking students who are involved in at least one blocking pair, show that the impossibilities remain, and characterize the possibility domain of priority profiles.