Nature's effect is finite and measurable... earthquakes are easily predictable in the long term... the terrible cost of radiation leak is easily predictable.
Like most other human efforts, the goal posts are pretty clear. Make nuke plants stronger than the strongest that nature is capable of producing.
A tsunami + earthquake in japan is not a surprise to anyone.
What is a surprise is that the people running the reactor consider it a surprise.
I wonder where "8.2" came from in the engineering spec. and whether shooting for the ability to withstand a quake somewhere in the neighborhood of, say, the 10th largest is the right target. Maybe "the largest + 200%" is a better place to start? Of course, it costs more money.
but that's not the fault of the plant's designers
But it is a totally avoidable error. Not the plant's designers error, but a human error nonetheless. Why are we limiting our concerns over human error to just the people responsible for the original design? Surely there were organizations/contractors involved in developing replacement generators. Surely there are current plant safety and maintenance directors. Surely there is a complete nuclear regulatory agency.
It is also likely that the money could be better spend elsewhere (that is, it could same more lives) such as in improving road safety or to hire more police officers.
Consider two possibilities.
One: design the reactor so that a magnitude-9 quake leaves it completely undamaged. Let's say that makes it cost $4B instead of $2B.
Two: design the reactor so that a magnitude-8.2 quake leaves it completely undamaged and a magnitude-9 quake (which might affect it, let's say, once per 200 years on average) makes it fail in a way that takes 100 lives and requires it to be completely rebuilt (to the same spec, let's say) at a cost of $2B.
Cost of option 1: $4B now. Nothing more. Cost of option 2: $2B now. $2B and 100 lives in another 200 years.
Suppose your discount parameter is 0.5% per year (much, much lower than just about any real person or institution uses). Then $2B + 100 lives in 200 years = $740M + 37 lives now.
Can you save 37 lives for $260M? Why yes, you can. Far more than 37 lives, in fact. If you do anything reasonable with $260M you may well save more than 37 lives just by (say) making everyone a little better off. (Wealth correlates with health.)
Therefore, option 2 is probably better.
Turning from hypothetical examples to the real world: $2B/$4B is probably an underestimate (in which case, the advantage of option 2 over option 1 is greater) and 100 lives is probably an overestimate (in which case, again the advantage of option 2 is greater).
On the other hand, the earthquake happened improbably early in the life of the plant. It is not an error to do things that work out well in almost all cases but badly when you get unlucky, when the alternative is something that works out (relatively) badly most of the time but well when you get unlucky.
Also, I'm not sure what you define as "failsafes", but the conventional measurement is number of containment levels. This reactor had 3, while I believe more modern reactors have 4.