If you're asking Server Fault for advice on this, you're definitely not on the leading edge of the hardware/systems side of low-latency design. That's fine, but this is a narrow space, and engineers in the industry tend not to share information.
I've managed trading systems in multiple environments, and the tuning parameters and tweaks made in one company almost never applied to other firms. Different application code, developers, frameworks, hardware choices, exchange and co-location options all combined to make it impossible to have a single formula for success.
The one constant throughout this was that there was repeated testing, refactoring and benchmarking involved. E.g. Try both settings and see. Make sure you have a way to measure at the application side. We can't tell you what will work for your environment.
I think I recall you using Windows for your application, so that's obviously a disadvantage since most people in this space have moved to Linux and nowadays, FPGA solutions for feed-handlers, filtering and strategy, 10GbE throughout, UDP and TCP kernel bypass, etc.
I don't think I've used an Intel adapter in finance for a few years. It's mostly been Myricom and Solarflare...
