Have you taken a look at standard deviation? Then you can take a variance simulator and check what "runs" you can expect over 400k hands with given winrate (assume 4.5bb/100) and given std.dev.