I ran it, and on my system (RTX 2070 Super, Ryzen 3700X) it yields an 8.7% performance difference in favor of the Sampler Feedback. That really is an enormous performance gain for what may be a (relatively speaking) easy to implement standardized feature, even if the real world benefits of it are half what they show in benchmark programs.