It is my understanding that many years ago NASDAQ imposed a many millisecond delay on connections based upon their proximity to the matching engines (they were located in Connecticut at that time, I think). The closer the end point of a connection was to their matching engines, the higher the delay. They did this to equalize the latency among all their connections (so West coast firms were not at geographical disadvantage compared to East coast firms).This only handles things up to the point of demarcation where they hand it off to the server itself. Then it becomes about who can use the fastest NICs, who can write the fastest software that's residing at the kernel layer or even no kernel at all, just full blown custom hardware.
Are you saying CME is utilizing some kind of complex per-switch QoS to guarantee a level playing field? I seriously doubt that. Sure they might be providing equidistant layer 1, but that's *just* layer 1.
I also understand that the SLA in Aurora limits the length of fiber between any customer cabinet and the CME's equipment to 12 inches. For data traveling at a rate of 10Gb, 12 inches represents about 1.5 nanoseconds.