The article says it better than me:
> [Banach-Tarski] means there can be no measure satisfying the requirements even when weakened from countable additivity to finite additivity.
Measure theory was an attempt to rescue us from the breakdown of the Riemann integral for poorly-behaved functions by adding a ton of abstraction of formalism. And it turns out that it isn't even successful at that.
But it turns out that what "poorly behaved functions" means is basically "non-computable functions", and so you can back up even further and point to that idea as the problematic one in the classical foundations of set theory -- that constructibility is not only a nice-to-have but is in fact necessary for any sort of coherent theory.
Put differently, Banach-Tarski shows that formalism-based systems that rely on the axiom of choice and the excluded middle too heavily result in behaviors which are clearly incorrect, but which are internally consistent.