If there are 500 measuring stations, then every year some one of them is going to experience a once-in-500-years event. We just never notice the other 499, and cherry-pick the one outlier after it already happened. (Yeah, there can be correlation between station measurements; obviously I'm speaking in generalities.)

Not really. You got it exactly backwards. The return period of an event is determined based on all recorded events. This does not mean that massive events that took place in the past didn't existed if no record was kept. Consequently, massive events are underrepresented both in magnitude and in frequency, and consequently estimates of their return period tend to be larger.

Think of it this way: the concept of the return period of an event (i.e., this is a one in X year's event) is determined based on the average time between similar events.