It is because of the higher memory latencies that Ryzen is currently saddled with compared to an equivalent Intel system. An Intel system runs with about 45ns latency compared to the best results so far of about 62ns on Ryzen. On release, they were at around 100ns so there has been some pretty big gains so far.
Physic scores have always been about where they should be, like cinebench, because by themselves they are not being limited by the frequency\latency performance ceiling. That all changes when you run both the GPU draw load and the CPU logic load together.
As the memory performance has improved, Combined scores that started at about 6500 at release day are now up into the 8000 range. 7700K will do about 9000 and a 6900K about 10000. Timpspy CPU scores were also at 6000 or so at launch and the best tuned memory systems now are now scoring 9000+ as the latency has come down.
The amazing thing is that the tech media in general is still basically in denial and goes out of its way to ignore memory latency at all costs. Maybe prattling on with the rubbish about windows scheduling and ccx thread switching is more glamorous?