I think the "zero" latency that Sanwa claims is due to the fact that anything faster than 2ms can't be done with normal digital servos. Therefore if you get a 2 - 2.2ms reading them you have virtually "zero" latency because you responding as fast as the servo can move.
Even with SR type servos the theoretical limit may still be somewhere between 1.5 - 2ms max for minimum latency. At that point I'm not even sure if a 380'ish Tx Framerate would be fast enough to get down to 1.5ms. Condac tested the 4PX in race mode with the TFhss Rx (telemetry off) to be at 2.1ms for minimum latency, so it's vertually as fast with digital servos as the M12S and MT44S. With that noted, the new SR protocol that Futaba is debuting for Sbus servos on the 7PX seems to be the same/similar to Sanwa's SR setup in that they can use these special servos that have a 750u. Therefore it may allow the potential lower latencies I've mentioned for the M12S/MT44S as well. It is interesting to ponder that the 4PX supposedly has the same CPU that the M12S, and upcoming 7PX use. The limiting factor for the 4PX's Tx frame rates getting towards the 380's Hrz of the M12S seems to be the Rx. Condac found that the 4PX's Tx frame rate actually changed based upon the Rx (Sfhss, Fasst, TFhss), topping off at 333 Hrz with the Tfhss (telemetry off). The upcoming Futaba SR protocol will probably raise this Tx frame rate. So we 4px owners are looking forward to hopefully having Futaba port this over to us as well.