There's been plenty of research for decades showing that 250ms response time in UX is a noticeable slowdown.
Except that there was a bug where if you go too quickly, some of the software safety interlocks didn't work right, and they had removed the hardware safeties in the update from Therac-20.
Most premium Android phones in 2016 were 120-30ms motion-photon response time (low to midrange probably hit 250), while iPhones were ~80ms. Haptics add another ~30ms delay. There have been massive improvements since 2016, as it has been a competitive metric, but even with >120Hz screens sub 60ms max true latency is uncommon.
As (actually my long term memory of) the MS video shows faster refresh and lower latency greatly improve smoothness and are useful for moving objects, especially scrolling (though tearing is a real issue). The problem is much more pronounced in VR/MR as eye tracking of diagonal motion is VERY discriminating. That's why they shoot for >240Hz refresh rates. Over 1kHz there doesn't seem to be a lot of improvement (2ms latency) due to a combination of cone/rod/neural filtering.
If you want a very generic worst case latency calculation based on the touch acquisition rate and display refresh rate assuming no position calculation delay it is:
Latency = 2/Ftouch + 2/Fdispl = 2/16Hz + 2/16Hz = 250ms
Please provide a link to your data and analysis.
Seriously, I hope you don't work for a major touchscreen supplier. A good UI can deal with relatively high latency input, a bad UI can make any latency unusable. Finger motion before/after contact and rapid high force contact bounces are common problems. Many high performance interfaces "gracefully" degrade report latency to similar levels under high interference conditions.
I'd also note that under video review people are notoriously bad at actually knowing how they interact with a touchscreen. Interestingly, fast response rates (and tactile response) reduce user complaints about proper operation even when UI errors remain constant. The users often assume they made the error rather than the machine. There are also interesting cognitive load effects.
You're saying people evaluate and react to checkbox feedback faster than 4 times/sec? I'm not a UX person so this claim seems laughable. I might believe twice that as beyond reasonable, but there is no way the average person waits less than a quarter of a second before assuming their input failed and duplicate the effort. One can hardly even observe and process that quickly - even when paying attention.