[if] variations in outcome are orders of magnitude greater than variations in valuation [...], then Graham's argument is mathematically obvious: get in to Youtube or Heroku at 1/10th the stake you'd get in Carsabi and you're still way ahead.
That assumes that variations in outcome are correlated with variations in valuation. If they are not (which may be the point of mentioning “black swans”), then an investor is better off investing at lower valuations.
That assumes that variations in outcome are correlated with variations in valuation. If they are not (which may be the point of mentioning “black swans”), then an investor is better off investing at lower valuations.