I spent a long time looking at an algorithm for long range force calculations called the Fast Multipole Method which is O(N) and found that practically it couldn't compete against an O(N log N) method we used that involved FFTs because the coefficient was so large that you'd need to simulate systems way bigger than is feasible in order for it to pay off because of cache locality, etc.
Honestly, that is not too uncommon. For top level algorithm choice analysis, "rounding" O(log(n)) to O(1), and O(n*log(n)) to O(n), is often pretty reasonable for algorithm comparison, even though it is strictly incorrect.
It is not uncommon for a simple O(n*log(n)) algorithm to beat out a more complex O(n) algorithm for realistic data sizes. If you "round" them to both O(n), then picking the simpler algorithm because the obvious choice, and can easily turn out to have better perf.
To be fair, in the general case (particles placed in arbitrary positions), it worked very well. The issue was that in the area of electromagnetics I worked on at the time, we make an assumption about things being placed on a regular grid, and that allows you to do the force calculation by convolving a kernel with the charges/dipole strength in Fourier space.
