Right, anything too expensive won't happen. And that's when you need to use some kind of objective metric to decide whether you can run the plant anyway.
If a nuclear plant has a probability of core damage of 1e-5 per reactor year, should we decommission it and build a new plant with 1e-7 probability? 100 times better is a lot, right? But if a core damage costs on average $100 billion, then the 1e-5 probability averages $1 million in disaster costs per year, and it's probably not worth it to decommission the plant and build a plant that will only average $10,000 per year. And it's DEFINITELY not worth it to decommission THAT plant and go for an even newer 1e-9 plant at $100/year in core damage costs. Somewhere, it just becomes good enough and it would be, in fact, irresponsible to add more safety (instead of going for extra road safety or something).
So where are we at? To my mind, we're quite good if we implement the cheapest lessons from Fukushima.