Would it make sense/be possible to re-implement/create alternative implementation for cluster that is based on Workers instead of child processes? Would make it possible to more efficiently pass data between the parent process and workers.