This paper presents improvements and extensions that have been applied to our earlier presented approach of mutually optimizing lithographic illumination and mask settings. Our work aims at two aspects: (1) improvements of the optimization approach and (2) of the simulation scheme used for the optimization. As described earlier, the main problem of the proposed optimization approach is the high requirement of computation time. One solution is to extensively distribute calculations onto different computers. As an alternative to the former approach using MPI, a new improved technique is proposed, which makes use of the Python powered framework Twisted. This allows for a fail-safe and load-balanced distribution of calculations in a heterogeneous network environment. Another enhancement is the integration of local optimization routines into the proposed concept. For that, the state-of-the-art optimization toolkit of Matlab has been integrated into our approach. By combining our genetic algorithm with local search methods it is not only possible to increase the overall optimization performance, but also to evaluate local environments in the search space, which helps to assess the technical stability of solutions. As a first example the non-linear SQP (sequential quadratic programming) has been used, allowing for constrained problem specifications. Furthermore, the simulation itself was drastically improved in terms of efficiency. For example, instead of evaluating all solutions at the same numerical resolution, as a first step, a coarse-grained evaluation is performed, only if a solution's merit lies above a certain threshold, a detailed analysis at a higher (numerical) resolution is conducted. Various tests demonstrate not only the increase in efficiency obtained with the newly incorporated measures, but also show new results for a combined optimization of different mask features.