Looked up the old discussion about quad view for future reference is someone is interested about the topic.
tldr; not good
The most efficient design would use variable rate, then encode in software right there in the gpu memory using the same map which was used for variable rate rendering.

Advantages
>no extra vertex shader work
>no processing power/transfer bandwidth wasted on occluded area
Disadvantage
>slightly higher gpu memory usage, you need big render target with high supersampled rate, even if most of the memory would use 1 or even less samples