Ok now we are here, 19posts later and you haven’t read my first post yet🤣
The simple explanation is that network driver is not perfectly optimized as in much more popular ramips 7621 platform. Sime day, maybe, but certainly not in next OpenWRT stable, it will get optimized IF someone takes care of making multi-core aware network queues/interrupts. For now it is running network on a single core forcing firewall offload on you to reach wire speeds.
OK, now I get it. Thank you for trying to help! I'll just go back to using offloading, which is perfectly fine for my use case.
I actually have, but not being a developer, I couldn't grasp the relevance, nor could I formulate an intelligent question to ask (subject matter too far out of the comfort zone). Now that @brada4 has explained it in terms that I can understand, I understand. 