Once upon a time in the 90's I was at work at 2am and I needed to implement a search over a data set. This function was going to be eventually called for every item, thus if I implemented it as a linear search, it would be n^2 behavior. Since it was so late and I was so tired, I marked it as something to fix later, and just did linear search.

Later that week, now that things were working, I profiled the n^2 search. The software controlled a piece of industrial test equipment, and the actual test process would take something around 4 hours to complete. Using the very worst case, far-beyond-reasonable data set, if I left the n^2 behavior in, would have added something like 6 seconds to that 4 hour runtime.

(Ultimately I fixed it anyways, but because it was easy, not because it mattered.)