That logic is reasonable if you can trivially afford 250k; in that case, you might choose to self-insure. However, that logic does not hold if the 1% event is not something you can afford.

Every dollar does not have the same incremental value. Going from $1B to $1B-$250k is not the same as going from $300k to $50k, and definitely not the same as going from $50k to -$200k.

That is why "and 250k loss would be more than 90 times worse" is there.

If loss of 250k would ruin my life and loss of 3k would be annoying, then paying makes sense.

Similarly, taking bet "double your wealth at 51% chance, lose every single thing you own at 49% chance" likely makes no sense, despite mean value being above zero.

> isn't it the same as saying that paying 3k for 1% chance of winning 250k is a good idea?

not in cases where loss of 250k would be more problematic than benefit of gaining 250k

for example loss of 250k may make someone homeless while gaining 250k would allow them to live a bit more comfortable - in such case bad effects of loss are much greater than gaining 250k

in other words, utility of money is not on a linear scale

It isn't a fallacy, it is plausibly optimal if the person being insured is involved in a non-ergodic process. The statistical argument you are applying is too basic to cover a large number of real-world situations.