Random Selection of Response Algorithms
Random selection of response algorithms lower the likelihood that a naive algorithm will be matched with a set of inputs that are problematic for it. The primary uses of this are:
This is a very useful technique for foiling an adversary, because when there is only one potential algorithm, it is relatively simple for an adversary to provide inputs that will make an algorithm perform at its worst. Random selection from a range of algorithms that perform the same task prevents this because different algorithms have different weaknesses, and an adversary won't know which to plan for.
It is valuable in hashing for similar reasons. There is no one hashing algorithm that is best for all inputs. A random selection among the possibilities helps reduce the chance that the selected hashing algorithm will be a poor match for the input.