This is a neat approach, Rob, and some form of it seems likely to be one of the best ways of thinking about this. I think the emphasis on putting yourself in the shoes of those you're trying to help rather than acting for yourself is particularly valuable. I think there is one extra difficulty that you haven't mentioned, though, which is to do with people having other preferences than yours.

Even if I'm able to work out that, given a random chance of being one of the participants I would prefer 2 to 3, it doesn't necessarily follow that 2 is preferable to 3 in an objective sense. It is interesting to imagine what the participants themselves would choose behind your veil (if they were fully informed about the tradeoffs etc.).

In many cases, one finds that people tend to think that their own condition is less bad than people who don't have the condition do. (That is, if you ask sighted people how bad it would be to be blind they say it would be much worse than blind people do when asked.) This suggests that, behind a veil of ignorance where self-interest is not at play, those at risk of malaria but not worms might regard treating worms as most important and those at risk of worms but not malaria would treat malaria. It seems hard to know whom to prioritise then.

There's also the eternal problem with imagining what one would choose - people often choose poorly. I assume you're making some sort of assumptions choosing under the best possible conditions. It may be, though, that your values depend on your decision-making conditions.

Of course, you still have to choose and like you say it's clear that 2 and 3 are both preferable to 1. I think this tool will get you answers most of the time, and can focus your mind on important questions, but there's a intrinsic uncertainty (or maybe indeterminateness) about the ordering.

I know you qualify this process as you own heuristic rather than a philosophical justification, but I fail to see the value of empathetic projection in this case which, in practice, is an invite for all sorts of biases. To state just two points: (i) imagining the experiential world of someone else isn't the same, or anywhere near to, experientially being someone else; (ii) it is not obvious that the imagined person's emotional and value set have any normative force as to what distributions we should favour in the world, i.e. X preferring Y to Z is not a normative argument for privileging Y over Z.

In Rawls' original position, judgement is exercised by a representative invested with a books-worth of qualifications as to why its conclusions are normatively important, i.e. Rawls tries to exactly model the person as free and equal in circumstances of fairness (it has frequently been argued, quite correctly, that Rawls' OP is superfluous to Rawls' actual argument, for the terms of agreement are well-defined outside of it). In the case of your procedure, judgement is exercised by whoever happens to be using it.

IMO, the possibility of normative interpersonal comparisons requires at least: (i) that we can justify a delimited bundle of goods as normatively prior to other goods; (ii) that those goods, within and between themselves, are formally commensurable; (iii) that we can produce a cardinal measure of those goods in the real-world; (iv) that we use that measure effectively to calculate correlations between the presence of those goods and the interventions in which we are interested; (v) that we complement this intervention efficacy with non-intervention variables, i.e. if intervention X yields 5 goods and intervention Y 10 goods, but we can deliver 2.5 X at the price of 1 Y in circumstance Z, then in circumstance Z we should prioritise X intervention.

I'm sure that, firstly, you know this better and more comprehensively than I, and secondly, that this process itself is a highly ineffective (i.e. resource-consuming) means of proceeding with interpersonal comparisons unless massively scaled. That said, I don't see why it shouldn't be a schematic ideal against which to exercise our non-ideal judgements. Your heuristic might roughly help (iii), and in this respect might be very helpful at the stage of first-evaluations, but there is more exacting means, and four other stages, besides.