Further optimization is of little value in terms of convergence. The problem is in the random-walk nature of the convergence. While the quotient is guaranteed to eventually converge, over a given interval the quotient can diverge. While convergence is guaranteed the time over which it converges is not guaranteed.