@feld It's around 6-7Gb/s. Better than I expected, but there seems to be some CPU utilization bug, because no matter how many threads I throw at it, all virtual CPUs will stay ~20% in the idle state. pf does not seem to make an impact after the first few runs have half the throughput for whatever reason. MTU also makes no difference, but that's probably because it's virtio.
image.png