Less is more -- the Dispatcher/ Executor principle for multi-task Reinforcement Learning