Even in the face of marginal utility differences, though, if you were to accept (unreasonably, of course; thought experiment disclaimer goes here) that you only had the two options given (do nothing and owe another $100 to the state, or take the bet, coming out $130 ahead or $230 behind), would you really suggest that doing nothing is the better option? How good would the odds need to be (and how little variance) for you to take the bet, if so?
If I laid out this payoff structure without placing it in a real world context where there are many other obvious potential solutions, I think you'd probably argue in favor of taking the bet - the expected loss is so much lower that I think it's difficult to make a marginal utility argument against it. I suspect that what's really making you argue against it in this example is that there are so many better ways to deal with the situation in real life that it's hard to even imagine having to make a choice between the two alternatives on the table.
I question this because you're already in the red: do nothing, and you'll be left with $130 in the bank but $250+$100=$350 in liabilities that eventually have to be paid, so you're at -$220 by default (even if you've still got $130 in your pocket, that's not really your money; as I commented elsewhere, I probably should have set up the example so that Mary had already purchased food for the week, to avoid questions of starvation).
When you're starting in negative territory the usual rules about marginal utility of money don't necessarily apply, I'd actually argue that there might be more utility gained when you go from -$50 to +$50 than there is going from -$150 to -$50 because of the debt spiral effects that you mention.
As for your other points, I agree, there are issues with the false dichotomy that I've set up, ones that I expected to be called out on. Perhaps I should have worked out a better example where there were fewer options available, c'est la vie.