On the stability of the Serrin problem