It's because the limiting factor isn't necessarily just the SSD's speed, it's the CPU and memory decompressing those assets after they're read, loading the GPU's memory from that, and the efficiency of the processes and hardware that move all that data around. Later on in the article they even talk about how many modern games duplicate commonly accessed data to compensate for the slow random read response of HDDs, which is an inefficiency easily dropped with NVMe drives.
I've had the chance to test load times between a standard ~520MB/s-read SATA3 SSD, and ~2GB/s and 3GB/s NVMe drives (Intel 750 & Samsung 960) in the same system, and the reduction in load times going from SSD to NMVe was maybe 10-15% despite 3-6x the read speed, but hardly noticeable between the NVMe's. On an OG fat PS4, swapping the internal ~60MB/s HDD for a SATAII-capped 300MB/s SSD cut power-on-to-Witcher-3 from three minutes to two. All that to say, faster storage helps, but it's far from a linear return, and until some methods free themselves from the shackles of HDD speeds (and random reads in particular), just adding raw speed isn't the solution.