Benchmarks are totally misleading in this case. Who cares if zip compression or rendering are a couple ms faster. You'll never notice that in real world tasks. Though, what you'll notice is that the OS prioritizes and sends off workloads to several cores when you're running several applications...
Most of my systems contain smoothed or modified data series, work with self organizing maps or other quantitative methods. Also, they all work on quote data rather than tick data.
I own a NT license and purchased QuantDeveloper source code. Initially I started developing systems with NT but as soon as you develop rather complex systems I hit it's limitations. Coding VB IDL is a pain ;) It's certainly nice for visual R&D but imho not the best for fully automated systems...