Because that's not what "the law of large numbers" is. LLN states the long-run averages of random events will approximate the expected value. Elections, states and voters are very non-random.
Yes this explains it, the states in question have tendencies that make them not at all representative of the country. NY and CA are way more urban and diverse. TX is also much more urban (or urban/suburban) but has other factors overriding this one.
Florida (historically) actually has some tendencies that made it likely to swing in tune with the country. The racial makeup is a decent cross-section and it at least has some population in rural, suburban, and urban areas. It’s still probably slightly too old and urban though, and (I think) its white/Hispanic/urban/suburban/rural populations are all more republican than the nationwide average for each group.
Theoretically a large state that is a representative sample of the entire country should be very close to the country’s results, but no state that fits this exactly. The transplant population of a state like Texas might be close, but it would still be too young/wealthy/educated to be a match