Reach higher bandwidths without bypassing the kernel

[HKalbasi]

I made a virtual interface using ip link add x1 type veth peer y1, and I'm able to get a 33Gbit/s bandwidth on it using iperf which uses tcp sockets of the kernel. But using tcpreplay -K -t I only get a 1Gbit/s bandwitdth. What is the difference between iperf and tcpreplay that makes so huge difference? Is it possible for tcpreplay to use advanced kernel apis like io_uring to reach higher bandwidths?