I've read the fireball algorithm and there was nowhere that I saw in which the hammer status has an impact on fireball behavior. Specifically, the probability of reversing direction appears to be fixed at 50% for each decision point and not dependent on hammer status or anything else. Similarly, the decision of whether or not to climb a ladder also appears to be fixed at 25% probability when it is moving right (but 0% when moving left) as long as taking the ladder is permitted but not forced.
There are still two possibilities in which hammer status could have an impact on fireball behavior. The first is the possibility that the programmers put the code for this behavior is some other area of the code outside of the fireball AI routine, which is unlikely. The second possibility is that there is a bug that causes hammer status to affect fireball behavior. While it is possible for such a bug to exist, in general behavior that is both complex and coherent is unlikely to be produced by a bug. In this case, the behavior described by both Dave and Dan is complex in the sense that the fireball behavior would require knowledge of where the fireball is, where Mario is, which direction is towards Mario and how long before the hammer expires and would need to affect each of the 5 fireballs individually in order to produce the appropriate behavior. The behavior is also coherent in the sense that it is behavior that might have feasibly been intentionally programmed in. So with this in mind I think you can have fairly high confidence that behavior like Dave and Dan describe does not exist.
