The advent of powerful digital signal processors has recently permitted the real-time implementation of model predictive control (MPC) in high-performance electric drives. Nevertheless, the use of MPC together with multiphase systems is increasingly challenging as the number of phases gets higher. On the positive side, the redundancy provided by the extra phases also opens the possibility to further optimize the control action. This paper describes the implementation of MPC for nine-phase drives using a three-step approach with an initial discarding of the switching states, a dynamic selection of the voltage vectors using hard constraints, and an improved performance including soft constraints. Experimental results confirm the ability of the proposed MPC to highly reduce the computational burden and switching frequency, while maintaining satisfactory steady state and dynamic performance.