Another tool for the non-confiscatory belt to consider is that the height of inputs is available at validation time. So one could, for example, except txn where all inputs are below some height to reduce the risk of breaking something.
One could also have a rule that sunsets and needs to be continually renewed to be maintained if you’re concerned about future risk. E.g. have it expire N years out and resoftfork in an extension every N/2 years. This does have some political risk from the repeated consensus changes but could still be superior if the alternative is doing nothing at all.
As far as the actual objective, it’s not clear to me that 1/10th is enough to make block construction sensible (though I’d be open to hearing an argument otherwise). … and pretty clear to me that much smaller than that would be a non-starter, not just because of the historical behavior but also because big transactions can be encountered in ordinary usage.