If the boxes are in a random jumble when the prisoner enters, there may be no practical way of having 100 prisoners consistently assign an ordering from 1 to 100 for the boxes. The approach outlined will then fail.
Were I a particularly evil warden, I could meticulously place all the boxes in the same such "jumble" after each prisoner left. One of the prisoners would certainly derive a different ordering of the boxes than one of the other prisoners. The approach might still have a better than random chance of succeeding, but it would be lower than that outlined in the article, since there are now two (or more) chances for a cycle of length greater than 50.