This is the heart of the issue - how many trade do you need to make to make a real statistically significant decision as to whether your system works or not? And not just works now, but worked last year, with different pairs etc.
Maybe 20 is enough, maybe if its a small edge you need 1000's...