On the constant in the tarry-escott problem