That was his point, actually. There is nothing wrong with reclaim, as your sample size gets larger your failure rate approaches the intended amount; in this case, 10%.

And FWIW, a 90% chance of success usually corresponds to a 10% chance of failure. When looking at a roll of the dice, you look at the percentage for success and you have to roll less than or equal to that on a d100. So a 90% chance of success means that if you roll a 1 to a 90 you succeed, and rolling a 91 to 100 means you fail.