Your hardware has been designed in 2014, when ~50 MBit/s were the kind of the hill. It's not fast enough to do more than ~200 MBit/s (and it's already with its back to the wall at those speeds, resulting in sawtooth like throughput patterns and delays). Software flow-offloading (hardware flow-offloading does not exist for this hardware) only goes this far, it's punching waaay beyond its limits.