Tested 4266 rgb,failed to train 4000 12 12 12 1.85-2V. Also latest 4266 were worser on tight timings compared to 3600c16. Latest 3733 c17 and 3600c15 again were worser,1707 serial week,1701 on pcb,644 week on IC.
On the average from 5 sets of 3600C16 i had way better average than on 7 sets of 4266.Not to mention that 4266 i had 3 kits with one module DOA.
It is what it is , i binned over 100 kits overall and i have 3 modules super good,found about 6 GOOD(4133 12 11 11 Spi 32M wazza) and about 10 4000-4040 which is in my book average.
My advice would be to get a few kits of 3600 17-18-18 or 16 16 16 NON-RGB and keep the best,return the rest.
Sorry to be the deal-breaker for the RGB but the different PCB makes handling a bunny, for instance lots of 55 codes at reboot or cold-startup at the same time when they were memtest-stable,this is PCB to blaim,might be additional noise caused by the leds or motherboards that are not perfect tuned for this.
I stated my own experience , you can believe me or not,this is opinion based as i said 100 kits of b-die G.skill retail tested,completely waste of cash ,time and resources . While you might see other "retail" kits 4200+ superpi stable, you know unicorns do not exist and real-life experience is so harsh