Reconfigurable flexible job shops are endowed with dynamic adaptability to respond rapidly to market changes. However, when machining workpieces with complex surfaces, completion of all operations in a single setup is infeasible; workpiece removal and secondary clamping are required, and their dependence on manual operations constrains the parallelism of clamping tasks, thereby reducing system flexibility. To address this, a mixed-integer linear programming model is formulated for the reconfigurable flexible job shop scheduling problem with secondary clamping, with makespan and the number of workers set as optimization objectives. An improved multi-objective genetic algorithm is developed. A three-vector encoding structure is adopted to handle complex constraints effectively. A hybrid greedy initialization strategy is designed to accelerate population convergence. Three neighborhood search strategies based on key operations are introduced to enhance local search capability. Validation on 15 benchmark instances confirms the effectiveness of the proposed operators, and comparison with four advanced algorithms demonstrates superior performance of the developed method. Application to an automotive differential manufacturing process shows that, relative to traditional scheduling schemes, the proposed approach reduces makespan by 56%.