Exactly, all those "mini divergences" eventually disappear

A miner that has accepted a newly invalid transaction into its memory pool and is trying to mine it, will keep producing invalid blocks forever until the owner shuts it down and upgrades. This was happening for weeks after P2SH triggered.

For instance, any miner that has modified/bypassed IsStandard() can do this, or any miner that accepts direct transaction submission, or any miner that runs an old node from before OP_NOPs were made non-standard.

On the other hand, the "single divergence" in the hardfork keeps growing forever (unless all miners evetually upgrade.

Which they do, because they will eventually notice they are burning money.

Sorry Jorge, but I don't think your argument makes sense.