Comment 8 for bug 217504

Hmm, looks like you got it.
I was starting to think the kernel was doing some kind of check to see if the key being read was the actual key being triggered, but I couldn't figure out how the kernel would know since all events go through the device file one way or another.

And it seems it was just as I thought. The change was committed in 2.6.23. Probably why no one has caught it up to now.