How do you account the traffic? iptables quota or connbytes? Note that packets in a connection which is offloaded won't go through iptables rules any more. luci-bwc and @jow 's nlbwmon still work, because they either use procfs or netlink to talk to the kernel and get the accounting information.
I'd have to doublecheck with nlbwmon, but I was looking at the counter in the firewall tab in luci. I actually only care about nlbwmon
Yep, just checked, both nlbwmon and realtime traffic in luci are dead silent (well mostly, just a couple of kilobytes) after enabling hardware offload. However, software offload indeed works.