Bottom line, if you have a beefy CPU with a bunch of bored cores (so any CPU with more than 4 cores for Arkham Knight) then Denuvo is likely to have little impact on your average FPS, while it does have a minor but noticeable impact on loading times and very much impacts initial game game startup time. There's also a minor increase in memory usage with Denuvo. No testing has been done with any hardware that has 4 cores or less (like i5 CPUs which are quite common still), as far as I know, and no testing has been done where the impact on the CPU has been measured either (as an alternative to using CPUs with less cores).
Article doesn't list hardware used for testing, that I could find, so I'm going to assume they tested on some top of the line system, which obviously is not representative of what most people actually use. They also don't say anything substantial about testing methodology This makes their conclusions essentially worthless, so we'll have to focus on the linked YouTube video (which had a Reddit thread that was quite interesting and included a lot of details about the testing methodology and its shortcomings, but at least gave us something tangible to work with).
Since most current day games are GPU limited, not CPU limited (barring certain strategy games, especially the turn based ones), using top-of-the-line hardware is likely not going to result in a noticeable difference in FPS, so if they don't actually measure CPU load it's hard to tell how much extra work the CPU needs, or does not need, to do (and it obviously will need to do extra work, encryption is not free). Testing on something that's more reasonable (eg. a quad core system) would be interesting to see since most games are incapable of using more than 4 cores (and afaiu Arkham Knight is among those), so a system where Denuvo actually has to share a core with the game would, imho, be a lot more representative and indicative of the effect it has (or doesn't have) than a system where there's a bunch of cores that are doing sod all anyway.
The tests in the YouTube video linked from the article video appear to have indicated that while the FPS differences were minor on midrange hardware they were there (which makes sense, given that on-the-fly decryption can't be a "free" operation and will become more detrimental the slower your CPU gets), moreover they indicated a massive decrease in initial game loading times without Denuvo (which isn't taken into account with the benchmark) and minor increases of in-game loading times. It should also be noted that he only ran tests 3 times and let the game "warm up" each time (iow, only the first run for each (Denuvo and non-Denuvo) was from a "cold" boot, while in successive runs data was already cached from prior runs, as indicated by the fact that only the first Denuvo run had the slow game startup). This kind of decreases the value of the results, small sample size and not rebooting between tests isn't ideal, I'd argue.
Whether 1-3 fps is significant is debatable, it probably isn't on a high-end system (then again, it might well be more than this on a lower end system), whether the extra CPU cycles amount to anything tangible has, as far as I can tell, been tested by nobody, so far.
There also wasn't any testing done with regards FPS stability (people noted on the video that the Denuvo version appeared to microstutter at points, can't say I noticed, but hey), if the Denuvo version results in less stable FPS then an average isn't necessarily going to really tell us much about the actual quality of the gameplay experience.
Whether the increased initial loading times (40 or so extra seconds, iirc) are significant is also debatable, and probably up to the individual. I know slow loading games irritate me (hi Destiny 2), but ymmv. The slower in-game loading times might be worrying to people on lower end systems since Arkham Knight streams data while playing, so quick movement around the game world might lead to more issues with Denuvo enabled (like there used to be at the game's release)