In what circumstances is it only 10x slower? Random search is totally useless when your environment is stochastic. These algorithms aren't learning sequences of actions, in fact most use a 30 'no op' random start to avoid just that.