Unfortunately for the risk of death idea due to the way that scales it means we should bring everyone to the point of almost certain death tomorrow to avoid one certain death right now.
I mean yes, the need to reduce values, emotions, and human life down to numbers is a known and justified challenge to a utilitarian ethic. But in many instances, it is generally allowable to do so, such as in the instance of taxation.
In that scenario, the utility value for the utility monster would have to be enormous. We're talking something that would probably also result in human deaths if it were to go unhandled, like triggering a war or nuclear meltdown.
I feel the greater ethical flaw in utilitarianism is the case where a utility value scales positively with suffering. What can you do if the utility monster derives positive utility value from others dying, as opposed to remaining linear? Utilitarian measures simply don't fare well when the actors are inherently antagonistic to one another.
There's no involvement of the utility monster in what I'm talking about.
If you put the cost of imminent certain death of one person as infinite, then given the choice between:
One person dies right now, everyone else survives and lives a full life.
And
In one hour, everyone in the world dies with a 99.9999% chance.
You would pick the latter. That's the issue I see with placing an infinite cost on imminent death.
If we're modelling utility in situations where human life could be endangered, you should probably tie some amount of utility value to risk of death (or inevitability of death, given the cookie example), preferably on a ramping scale that approaches infinity as death becomes more imminent.
Maybe I'm missing the forest for the trees here, maybe the point is that it's impossible to create a moral system for allocating resources in a world where they are finite. I see little utility in such an argument, however.