IMHO this is pure nonsense unless you don't measure it in a lab, with some instrumentation.
Let's see what happens, in a very simplified scheme:
- Your finger presses on the key. Maybe you can consider it obvious, but you have to command it to move and feel the physical feedback. This takes time, even for the feedback alone it's in the milliseconds' range.
Say for simplicity that you're doing this part with a robotic finger and that you record a timestamp of the event
- The keyboard controller scans the keys and discovers your key's pressed. Fine. This takes time, depending on the keyboard connection matrix, but if you're with a fast enough device maybe it's just microseconds (there are some keyboards sold to have no matrix scan, fast key reads, etc.). Oh well, not considering the debounce algorithms... because pushing and releasing are two well-defined states
- Then the keycode has to be sent to the PC, which is a variable timing due to bus/line/channel operation - USB speed, controller operation; if with a radio channel, there's a possible delay due to radio interference.
- Now is the time for the BIOS/device driver/OS queueing and de-queueing of the event, sending it to an application that registered itself for receiving it.
- The application will then have to do something with it, at its own pace.
Anything that any simple software alone can do is on points 4-5, while for the first three, you have no way to measure the impact. Unless, as I said, you timestamp every event with an independent external instrument, that can correlate all the steps.