Chrome/FF in this case has an UI thread that becomes unresponsive for whatever reason and doesn't empty it's message-queue (that contains the mouse/keyboard/etc events).
In the case of IntelliJ there is probably some of their code accepting the event and then probably has some ad-hoc message passing by setting a field that is polled by the receiver thread under the assumption that the receiver always runs faster than a human can act, however as soon as that assumption is broken by some pause you get what you describe.
In intelliJ it was like whenever the UI was not immediately responding to mouse events, it would often just lose the event instead of it being correctly processed once the UI thread was freed up.