When everything that can go wrong, does. This is a popular concept in computer science, especially with algorithms. For example: If you create a binary search tree by iterating through a sorted list, you'll get a "tree" with no branches, which is worse than useless. See run-length encoding for another example.

Bubble sort is an interesting algorithm because the best case is just as bad as the worst case, and so are all the other cases in between. Bubble sort is taught to CS freshmen because it's not only inefficient nearly to the point of uselessness, but also relatively unfathomable in comparison to other more efficient sorts. It's still better than bogo-sort, if you're into faint praise.

This is also an important concept in Electrical Engineering. Instead of a component such as a transistor or resistor having known attributes its attributes will be confined to a certain tolerance. It is often desirable to design your system so that it will perform correctly if the worst case of tolerances occur.