Let's see.
https://web-backend.simula.no/sites/default/files/publicatio...
> At the most sensitive, our findings reveal that some perceive delays below 40 ms. However, the median threshold suggests that motorvisual delays are more likely than not to go undetected below 51-90 ms.
By this study's numbers, 20ms is somewhat below the lower limit of ~40ms, but not too far below. 100ms would be easily perceivable - though, based on the other replies, it seems that VS Code does not actually have that much latency.
Don't confuse this with human reaction time, which is indeed an order of magnitude higher, at over 200ms. For one thing, reaction time is based on unpredictable events, whereas the appearance of keystrokes is highly predictable. It's based on the user's own keypresses, which a touch typer will usually have subconsciously planned (via muscle memory) several characters in advance. So the user will also be subconsciously predicting when the text will appear, and can notice if the timing is off. Also, even when it comes to unpredictable events, humans can discern, after the fact, the time difference between two previous sensory inputs (e.g. between feeling a keyboard key press down and seeing the character on screen), for much shorter time differences than the reaction time.
Of course, just because these levels of latency are perceptible doesn't mean they're a material obstacle to getting work done. As a relatively latency-sensitive person, I'm not sure whether they're a material obstacle. I just think they're annoying. Higher levels of latency (in the hundreds of ms) can definitely get in the way though, especially when the latency is variable (like SSH over cellular connections).