Assuming that a meteor impact is totally random (which might not be true because objects that impact the earth could conceivably influence each other through gravity, etc), it is probably best to model these impacts as a poisson process.
A poisson process with an average arrival time of one million years means that we would find that if these size impacts happen on average once per million years that on any given day, like the day before this Greenland discovery, we should expect the next impact to happen a million years in the future and the previous impact to have occurred one million years in the past.
If buses arrive randomly and independently on average every 10 minute, then we can expect a ten minute wait at the bus stop if we get there at a random time.
Think of random points on the number line averaging a point every million years. Now throw a stochastic dart at the line. It’s much more likely to land in a large interval not the small ones—-they take up more space on the line. For this reason, the distance between the dart and the next random point is on average one million, not 500,000. So it is very unlikely that the meteor impact was only a few thousand years ago.