Someone who donates both to the mosquito nets effort in Africa, and to the society which helps stray dogs and cats in Michigan, is not necessarily being irrational. They just may be perceiving the two benefits to lie on incomparable axes.
If you decide that donating 1$ to mosquito nets and 1$ to stray dogs is better than 2$ to one or 2$ to the other, then you have in fact performed a comparison between those three actions. If the type of good generated by mosquito nets is one axis and the type of good generated by saving stray dogs is another, then the scalar-valuedness of utility isn't about the axes, it's about comparing any given point in that 2-D space with any other point.
The alternative to being able to compare things isn't some decision process other than comparison. The alternative is to not have preferences about the state of the world at all; to say that there is no such thing as a "right thing to do" in a given circumstance.
Why not all other activity?
Expected utility does apply to all activity.
then... it's about comparing any given point in that 2-D space with any other point.
Granted, preferring one particular 2D point to another may be read as running a scalar-valued comparison function on the 2D space (such a reading is not without problems, e.g. because real people's preferences may not be transitive, but let's ignore those details). However, from the existence of such a function it does not follow that "we should find the most efficient group and give it our entire charity budget" - this being the claim the universality of which I was contesting.
If it's worth saying, but not worth its own post, even in Discussion, it goes here.