So let's take another example: the gene pool of any given species. It's about resilience too.
Not useful enough genes exit the gene pool regularly, does that mean that the gene pool of any given species end up gravitating toward a single gene?
If low bandwitch nodes exit the nood pole that only means we will have a fitter node pool. It doesn't imply anything about the node count.
The issue is with the pace at which an uncontained block size would allow for large entities to capture the network.
Your analogies are all pretty good but they're beside the point.
Following your argument most of the nodes composing the network as we speak would vanish and would quickly turned into a specialized market likely to consolidate into few large entities.
One of the famous false prophecy of Karl Marx is that the free market mechanism would be lead to the monopoly of few corporations (thanks to economy of scale) that would end up controling the world production.
In other world, he tought unconstrained competition would allow for large entities to capture the economy.
The fact that "no constraint" lead to "capture of the stronger" is a very intuitive and very widespread biais, but it's not something which have a strong empirical basis.
On the other hand, we know that protecting a system from the stress of competition turn out to be desastrous for the long-term survival of the said system, because prioritizing the protection of the weakest points of the system lead to fragility of the whole system. Confer Taleb's
Antifragility book for an expanded version of that argument.
In a world without blocksize, we don't know how many nodes would vanish nor how many new nodes would appear, but we can be pretty sure that a lot of the existing nodes would upgrade, and thus the nodes network would be more resilient. Focusing only on the nodes that would vanished, without taking into account the creation of new nodes and the widespread improvement of preexisting nodes is a partial view of the problem.
Following the logic of protecting the network from competition to preserve the weakest points we should had limit the increase of mining difficulty to avoid mining specialization (it's obvious that the network would have been far less resilient).