I’m not sure I understand what you want. At the lowest level (the physical cable running between the server and your switch) the network card is going to be acting in a manner that could be seen as ‘single-threaded’.
Specifically, it’s going to be dispatching a series of IP packets in a spray-and-pray fashion, and the higher level (TCP) will take care of detecting loss and re-dispatch.
The key indicators of performance you guys should be looking at are throughput (i.e., MB/s sent over the network card) and latency (average and max - the max latency is where Java enters into a GC cycle and underperforms). These need to be measured by a third-party widget (not the JVM you’re running your code on).
Going back to real-time systems, you’ll also want to measure jitter. That’s essentially deviation from the average (i.e., the difference between your min/max and your average latency above).
In the worst case, perhaps you guys should fork Netty and make the changes there, not that I understand the change particularly well. At least then you’ll have a reasonable starting point instead of building it from scratch (know that thousands of hours have gone into Netty - that’s hundreds of thousands of dollars in salaries for a business).