I thought they vastly improved user-space wireguard performance?

https://tailscale.com/blog/more-throughput

Not sure if the kernel implementation pulled ahead again, I don't really follow these things.

Also not defending tailscale, I respect them but I agree they are a one size fits some solution.