This work addresses the design and experimental implementation in real-time of an integrated predictive load-split management system for the transient and fluctuating propeller load sharing. Control-oriented modeling of the power system was performed based on experimental data gathered from the hybrid plant and on first principles for the diesel engine behavior and battery charging. Propulsion plant and environmental disturbance models are developed to simulate realistic marine load application. A nonlinear model predictive control (NMPC) scheme is proposed for the optimal transient power-split problem of a hybrid diesel-electric marine propulsion plant. The NMPC scheme directly controls the torque output of the diesel engine and the electric motor/generator ensuring that certain constraints concerning the system overloading are met, avoiding fast accelerations and load fluctuations of the diesel engine that affect engine performance. To achieve offset-free model predictive control (MPC) control, an observer is developed to provide the propeller law parameter to the NMPC for load estimation. The control system was experimentally tested in real-time operation. Results showed that controller rejected load disturbances and maintained the desired rotational speed of the powertrain as well as the desirable state of charge (SOC) in battery within the power plant limits, achieving smooth power transitions and mitigation of power fluctuations of the diesel engine.