Right, if we consider O(n) "ideal" (for many domains, this is true), we can consider processing per item as a more convenient metric. If you go above, say, O(log n) processing per item, you're in trouble proportionate to the unintended success of this portion of the code. Very easy to write a sloppy utility with a nice API, only for it to be 'the perfect fit' for someone else's tight deadline, become an idiom in the code, finally see a 10,000 entry example much later