Polymerization processes exhibit complex dynamics where achieving precise control is essential. This paper compares various predictive control strategies for a multiple-input, multiple-output system. The studied system includes two plants: a nonlinear kinetic model of a propylene polymerization reactor and a linear model of a counter-current shell-and-tube heat exchanger for removing heat from the exothermic reaction. The controlled variables are the reactor temperature and melt flow index, controlled by coolant and product removal flow rates. We evaluate optimal control strategies based on model predictive controllers (MPC) with differing model structures: (i) nonlinear, (ii) linear, and (iii) linear multi-model approaches. The multi-model strategy uses a weighted predicted output, calculated by combining model outputs using weights derived from prediction errors. Unlike traditional multi-model controllers that constrain only the weighted output, our method enforces constraints on all individual model outputs. This enhances robustness, leading to reduced output oscillations, shorter settling times, and fewer constraint violations. Additionally, multi-model controllers maintained computational efficiency and achieved objective function values within 5% of the nonlinear MPC.