If it was a CPU bottleneck then why are the single cards so much faster. They would have the same CPU bottleneck.
It makes no sense that under DX11 1080p is slower than a single card but in DX12 it's the 4k resolution that is slower than a single card. In DX11 the 4k resolution is 50% or more faster than a single card, but in DX12 it's 1080p that is nearly 50% faster than a single card.
The only logical reason I can think of for that is that the drivers are broken. And if we assume that the drivers are broken then we should probably assume that none of the benchmarks are very reliable.