...you know, it's only after I read this comment that I realised that you're suggesting that the utility-decreasing aspects may not use the same function as the utility-decreasing aspects. That is, what I was doing was mathematically equivalent to first linearly combining the separate aspects, and only then feeding that single number to a monotonically increasing nonlinear function.

Now I feel somewhat silly.

But yes, now I see that you are right. There are possible ethical models (example: bounded asymptotic increase for positive utility, unbounded linear decrease for negative utility) wherein a larger Omelas could be worse than a smaller Omelas, above some critical maximum size. In fact, there are some functions wherein an Omelas of size X could have positive utility, while an Omelas of size Y (with Y>X) could have negative utility.