If every single 3950X they tested can maintain 4.00GHz all-core under their testing criteria, and a majority of them can hit 4.10GHz, it seems a bit silly to pay a $100 premium for the bottom-of-the-barrel, lowest-8% of chips when there's at least a 92% chance that one bought at random will be better. I suppose the 3950X is hard to come by at the moment, and markups by small sellers are to be expected, but just to get a guaranteed average-performing chip, they charge a $400 premium. I guess if they didn't, 3rd-party resellers might buy them up and do the same though.
Also, with a less than 4% all-core overclocking difference separating the best-performing chips from the worst-performing ones, it all seems a bit pointless. And going by their data, chances are that a 3950X bought at random will manage 4.10 GHz under their test criteria, so the best-performing models they sell only manage a little over 1% higher clocks than average, but cost twice as much as MSRP.
And performing an all-core "overclock" on the 3950X seems a bit pointless in general, unless perhaps a system is only used for heavily-threaded workloads that utilize all cores, and lightly-threaded performance doesn't matter. Otherwise, you're trading a notable hit to lightly-threaded performance for minimal gains to heavily-threaded performance.