I believe it's referencing this: https://github.com/KonnorRogers/view-layer-benchmarks
There were a few things off about the benchmarks so I made a pull request with some updates. The 12x claim is WAY off. https://github.com/KonnorRogers/view-layer-benchmarks/pull/2...
Can't believe this still has to be debunked. No Phlex certainly isn't 12x faster than ERB. It's noticeably slower than ERB, e.g.: https://github.com/KonnorRogers/view-layer-benchmarks/commit...
In some extreme scenarios with tons of very small partials, it can win against Action View because the Action View partial lookup is significant overhead.