The linked document concerning performance of local variables does not take into account that we here with PSoCs have different architectures. Take the PSoC3 where we have a lot (more than 2) of different memory areas with different access models and the situation that the stack is not usable for local var allocation. Here we have (again: "normally") no difference between global and local vars. Looking at the PSoC1 core which comes with just two registers there IS a difference between the performance of bank0-, bankN vars and locals. PSoC5 has got a modern core where there are no differences between memory accesses on the stack or in SRam.
Bob