Which is exactly what Mellanox and others are doing. At QuasarDB, we’re using their drivers to bypass the kernel entirely, which (at scale) saves tremendous amounts of CPU load.
[1] http://www.mellanox.com/page/products_dyn?product_family=209...
100GBit would usually put enormous strains on the CPU (to the point it’s pretty much only doing that), which is taken away entirely by using this.
As it turns out, hardware appears to be much faster at these types of things than general purpose CPUs. :)