Or they mean that the game is done in 2/3rds of its 16.6 per frame rendering budget, and use the remainder to buffer input, predict / correct for latency etc. Which makes me wonder - in theory, if you can afford it, would it make sense for a fighting game to wait for the first 6.6ms (maybe doing networking stuff) and then take the controller state and render the frame immediately afterwards? Would that give a faster response to controller input? Otherwise the minimum response time would be between 16.6 and 32.2 (one frame detecting the controller input, next frame renders the new graphics), but now it could go as low as 10ms with 26.6 as maximum? Does that make sense at all?