Simulation-optimization problems exhibit substantial inefficiencies when applied to high-dimensional problems. The problem is exacerbated in case where feasibility also needs to be evaluated using simulation. In this work, we propose an approximate iterative approach to identify feasible solutions and quickly find good solutions to the original problem. The approach is based on discrete event optimization (i.e., a mathematical programming representation of the simulation-optimization problems) and Benders decomposition, which is used for cut generation while a system alternative is simulated. The procedure is currently tailored for the server allocation problem in the multi-stage serial-parallel manufacturing line constrained to a target system time on a specific sample path. Results on randomly generated instances show its effectiveness in quickly eliminating infeasible solutions, thus decreasing the required computational effort and keeping the optimality gap low.