@benroyce @numodular @georgetakei
You're talking about a single round of elections. The claim is about repeated rounds.
Let's play through your example:
After losing to red, what does the blue team do? They can move their policies to be more red or more green. Duverger predicts that all of the parties will move to some centre point, and people will abandon the green team because the red and blue teams are close to some average.
The problem with this is that it leaves an increasing number of people disenfranchised. So it's easy for voters on the bluer or redder end to look at the blues and reds and not be able to tell the difference between them. At this point, a green party can start peeling off a lot of votes from either the red or blue party, depending on what policies it goes after.
The big thing that Duverger's Law misses is that the red team isn't just competing against the blue team, they're competing against apathy. If the red team moves to some arbitrary middle and the blue team doesn't, the red team may have more support, but the blue team will have more enthusiastic support. You don't win an election because people like your platform, you win an election because people think your platform is worth bothering to turn up and vote for (compulsory voting changes this slightly).
That green party from your example isn't taking voters from the red or blue party, they're picking up people who don't like the red or blue team enough to bother to vote for them.
The red and blue team can pick up some apathetic voters by scaring them: vote for us or the other team wins. But that works only in the short term if the reds and blues are moving together. After a while, voters don't care.
Once you have 30% of the voters in the apathy box, you're in a very unstable situation: a third party that can pick up a lot of those voters can start winning.
The reason this rarely happens in the USA is that third parties keep aiming at the highest-risk races first. A third party with no track record stands almost no chance of winning the US Presidential election, but that's not where you start. You start in local elections. Then you get a track record in city councils and so on. Then you go for state senates and legislatures. Then state governorships. Then for the US house and senate. Then you aim for President. And that takes a long time, and will be slower in a country the size of the USA.