Are there examples of whether each of these implementations provide a better performance for a certain kind of computation over others?

If the interconnect occurs between the cores' local caches, does that mean that we will always be working at the speed of the shared cache if we access variables from programs running on different cores.

I have two quick questions about the interconnects. The being what is the significance of two blocks of memory on the interconnect. Are they physically separate, or just two different access points to the same physical memory? Secondly what is the function of the nodes the multistage network. I presume they fulfil a switching function and if so what sort of information are they making their decisions based on?