The Big-O notation doesn't work in such a simple way as taking the complexity of two algorithms and comparing them directly like that. Big-O notation relies heavily on exactly what 'N' is. Just because two algorithms are O(N) or O(N^2) doesn't mean a thing, because in one algorithm, N could be a large, time-consuming mathematical calculation, and in the other N could be a simple regular expression.

The power of Big-O notation comes with being able to roughly be able to predict how an algorithm will act on different sizes of datasets, and taking that information and tailoring your algorithm to get the best performance based on the fastest running time of the average-sized set of data. If one algorithm is O(N), and the other is O(N^2), the latter may be a better choice in some cases, if the former has a much larger N, and you can ensure that there won't be too much data to negate the smaller N.

Basically, it all boils down to the necessity of doing much more testing than you did. The best efficiency takes many steps to reach. If you did more benchmarks with different numbers of IP addresses, different complexity algorithms, and things like that, you'd start to see how the Big-O notation would help you predict future tests.