I've been running two GPUs - 6800 in slot 1, Vega 64 in slot 3 - for the last month and it's working fine. In macOS I use an SSDT to disable the 6800, and in Windows I use "Disable Driver" to disable the Vega 64.
I did briefly try running Windows with the Vega 64 enabled, but I started getting lower frame rates and jerky action in the game I was playing. Not 100% certain it was connected to enabling the Vega 64, but I disabled it again to be sure. I haven't connected that extra power cable for running two GPUs, which I guess I should if I plan to have them both enabled. I did wonder whether maybe the 6800 wasn't able to draw as much power with the Vega 64 enabled and that caused it to run slower, though I'm not sure if it works like that; I'd have expected a crash or blue screen if I had power issues, not lower performance.
I also run DaVinci Resolve and saw that Puget review and it was quite disappointing. However, as you mention, I took solace in the fact that they tested on Windows using OpenCL. Windows Resolve doesn't use AMD's Vulkan, and OpenCL is known to be much weaker than NVidia's CUDA.
I believe macOS' metal is regarded as a pretty capable and powerful compute API; potentially a rival to CUDA, albeit much newer. Therefore my hope is that Metal + AMD will be much more competitive compared to CUDA + NVidia than OpenCL + AMD is.
When the 6000 series GPUs are supported in macOS I plan to do a benchmark test, comparing Resolve in Windows using OpenCL vs Resolve in macOS using Metal. My hope is that I see noticeably better performance in macOS. I don't have an NVidia GPU to compare against, but if macOS Metal does outperform Windows OpenCL then that would be some indication that that Puget comparison benchmark isn't applicable to video editing on macOS.
Regardless of API, once fully supported in macOS I'm sure 6000 series GPU will outperform a Radeon VII.