What you really want to benchmark is how this ARM system (and especially a number of them working together) fares doing several typical server tasks, and compare that to other systems/clusters doing the same. And take TCO, including space savings & power consumption, into account while doing so.
Because clustering a whole bunch of these to work together is what they are supposedly good at:
That's an on-chip network switch with 8 ports (3 internal, 5 external) that can handle a total capacity of up to 80 Gbit/s, or up to 10 Gbit/s per channel, which can also automatically slow down (to save energy) depending on the current requirements. To compare: a standalone ethernet switch which can handle speeds like that would consume more than 5W on its own (and also require a lot more space, of course).