I'd say that the solutions that the discussed algorithm comes up with are, on average, 1.37% worse than the optimal solution. I.e. you are not solving the problem, but you're damn close.