If your theory is correct highly complex games like Chess, Dota, and Go this shouldn't possible. Yet we've seen it happen.
All that has to take place is clustering, not optimal clustering just clustering.
We're also probably attacking the problem from different angles - I'm mostly thinking about completely disregarded strats/characters.
For example in Dota push strategies (where you try to overwhelm your opponents very early in the game, a sort of all-in strategy) were valid during a very short period, one of about 6-9 months in 2008, IIRC. Then they were basically impossible until 2012.
Also some heroes have been constantly bottom tier ever since competitive Dota started in ~2006 (Spirit Breaker - I'm looking at you!).
It's much more easier to experiment as a single player. Once you get more people involved, social dynamics start happening. And there's nothing harder to budge than social inertia, IMO.
Look how long it took for some of the characters presented in this game to surface (6-8 years for some)- that cycle could literally take decades for some heroes in Dota & co. I actually like patches as a forcing function in this case.
Your examples are actually great at proving my point: it took decades if not centuries for the current varied chess and go strategies to emerge. By that point we won't care about Dota & co anymore, we'll all be in the Matrix :)