Your result is not only the time your nervous system took to see and respond. It includes the time needed for the stimulus to reach the screen and the time needed for your input to travel back through the device, operating system and browser.
Calibrated reaction-time research has shown that hardware/software delays can meaningfully inflate measured reaction times. Woods et al. measured an 11 ms display delay on one LCD setup and corrected their reaction results for hardware. PC-PVT researchers similarly noted that general-purpose computers can introduce timing uncertainty unless the system is characterised carefully.
This does not make browser tests useless. It means the strongest use case is relative measurement: same device, same browser, same test, repeated over time.
Reaction Lab timestamps responses in the browser with performance.now(), uses random delays so the signal is harder to predict, and stores personal bests locally where appropriate. It still cannot remove your device's physical display/input latency.
Continue with: Input lag vs reaction time · Mobile vs desktop · How many reaction trials? · Why scores vary.