I don’t have any references to scientific papers or such to share, but I can speak from my own experience: monomodal gaze interaction used for e.g. controlling the mouse pointer leads to unnatural eye movements and eye strain. It gets better with time: I don’t know if it’s because the eye muscles adapt, or because you get more used to it and relaxed after a while — ironically the gaze interaction works better the less you think about it — or both. Probably both.
For your second question: no, it’s not the same filter. The IV-T algorithm described in the white paper is for post-hoc analysis and makes use of both past and future data points. The EyeX Engine uses a real-time algorithm based only on the current and the past.