As emerged from the discussion in comments to my original post, the following claim of Hare's is very plausible: For any consecutive worlds Wi, Wi+1 in an up-slope morphing sequence, minimal benevolence requires that an agent "actively favor" the latter, where this means that were they to become agnostic about which world is actual, and they could choose to actualize either of these two, they would choose to actualize the latter. Note that "actively favoring" Wi+1 is compatible with actually preferring Wi, because the agent's actual preference might be contingent on their modal knowledge that some other world (WJack, in the previous discussion) is actual. "Active favoring" is a kind of counterfactual preference, or disposition to prefer, in a special circumstance of choice. If your actual circumstances differ from the above described special circumstances, then it's no surprise that your actual preferences may likewise differ.
Ok, hopefully that's all clear enough so far. Now, does it follow from this minimal claim that we must prefer the final world in an upslope morphing sequence to the first? Not immediately. So far we are merely committed to preferring each Wi+1 to Wi when in a position to actualize either. But in a typical case, our only options are to bring about either extreme case -- WJack or WJill, say -- the intermediate worlds in the morphing sequence are not accessible options for the agent. We thus need a further principle of rationality, what Hare ('The Ethics of Morphing' [pdf], p.20; p.122 in the published version) calls The Practical Insigniﬁcance of Irrelevant Alternatives:
If, given the options of bringing about S1, S2, S3, you will willingly bring about S3, then given the options of bringing about S1*, S3* (complete states of affairs relevantly just like S1 and S3, but in which the option of bringing about S2 is not available to you), you will willingly bring about S3.*
This is an attractive-sounding principle, but I think Hare's morphing argument can be understood as showing why a partialist should reject it. The sort of partialist I have in mind is one who accepts the 'person-affecting view' that we should care about particular individuals, rather than humanity in general. But they can't plausibly hold that only actually-existing people matter, such that the moral status of an option depends on whether or not it's the one you actually choose. For example, the partialist should not say that, faced with a choice to bring into existence either Jack or Jill, I should discount Jill's welfare because I'm actually going to choose Jack. Rather, I think, they should hold that you ought to have a particular concern for each candidate for being actualized -- but not for other merely possible individuals that aren't candidates for actuality.
To illustrate, suppose we have a 4-world upslope morphing sequence from Jack to Jill through Baby1 (who is a better-off counterpart of Jack) and Baby2 (who is a worse-off counterpart of Jill, and a better-off counterpart of Baby1). My suggestion is that the partialist's preference ordering over these alternatives will depend on which of them are live candidates for actuality. Let's compare two cases:
(1) All options on the table: There are four buttons before us, one for each world as described above, and whichever button we press first will cause the corresponding world to be actualized. Because Jack, Baby1, Baby2, and Jill are all candidates for actuality, minimal decency requires us to prefer pareto-superior alternatives for each. So we must prefer W1 to WJack (for Jack's sake), W2 to W1 (for Baby1's sake), and WJill to W2 (for Baby2's sake). We are thus committed to actualizing WJill.
(2) Only WJack and WJill are options: In this case, minimal decency merely commits us to preferring W1 over WJack, and WJill over W2; we are not required to prefer W2 to W1, since there is no candidate person (viz. Jack or Jill) for whose sake we must have this preference. It is thus rationally permissible to actualize either WJill or WJack.
In short: The middle options in a morphing sequence, despite being unchosen (or dispreferred to the final option) are not irrelevant alternatives. Their moral relevance consists in their introducing additional candidate individuals that the partialist is committed to respecting (by preferring options that are pareto superior for these individuals). In the absence of any such live option, the mere existence of these intermediate (but unaccessible) possible worlds will be of little interest to the partialist: They are not committed to preferring that things be better for these unaccessible people; they may prefer that a (less well off) counterpart of a modally accessible person exist instead.