I believe your analysis is ignoring the convexity of the NPV vs stochastic survival rates. If survival is not assured (a reasonable assumption), there is additional benefit to early withdrawals for which you are not accounting.