It is about min-maxing. If you have backend written in Rust which uses Hyper, for example, Caddy will be bottleneck.

Depending of course from the type of the backend (is it limited by other I/O and Caddy bottleneck does not matter)

I've never used ferron, but if you look at the graphs, he gives comparisons.

So, I guess, performance + easy of use. Obviously, caddy is much more mature though.