Evaluative terms like 'instrumentally good' (or 'bad') are global: they may be applied to any kind of thing -- not just acts, but also rules, motives, and even such 'mere evaluands' (utterly disconnected from rational agency) as climates and eye colours. This may lead us to wonder: are all normative terms likewise 'global' in their application, or are there some important normative properties (e.g., perhaps, moral rightness) that apply only to agential features, and not to mere evaluands? The former view -- what we might call the parity thesis -- is what I think sets global consequentialists apart from ordinary act consequentialists (if anything does). Here I want to suggest some grounds for rejecting the parity thesis, and instead holding that there's a normatively significant difference between 'agential' and 'mere' evaluands.
We may begin by noting the intuitive strangeness of talk of "the right eye colour to have been born with", or "reasons for eye colours", in contrast to "right action" and "reasons for action". Of course, merely sounding strange is not conclusive evidence of anything, but it should at least raise our suspicions. So let me now try to support these suspicions by noting some more principled points of difference between acts and eye colours.
* Eye colours aren’t judgment-sensitive, and so cannot be possessed for (normative) reasons, the way that one can act for reasons. Judging that it would be better to have blue eyes will not even tend to result in my having blue eyes -- at least, not as a direct response to my rational judgment. I may, of course, judge that I have reason to act so as to change my eye colour, and in response to this judgment perform that action. But this is a case of acting for a reason, and not - in the first instance - a case of "possessing an eye colour for a reason". To support this, note that the fact that the eyes in question are mine plays no essential role in the story. I might just as well have acted to change someone else's eye colour. So the reason here is clearly one for the acting agent, not the eye-colour-possessing subject (who may not be an agent at all). What the reason elicits is an act, not an eye colour, from the reasons-responsive agent.
* Agents are only subject to normative criticism for their agential failures; not for merely physical flaws/disutility. I mean, you might call someone 'ugly', but that's clearly not a normative (moral or rational) criticism in the way that it would be to call their actions selfish, or their beliefs inconsistent. It wouldn't make sense to demand that someone justify their ugliness, the way that one can be asked to justify their beliefs and actions.
Even if it would've somehow had great instrumental value were I to have been born with blue eyes (insert trickster demon story), the fact that my eyes aren't the lucky colour wouldn't seem to reflect poorly on me at all. It'd just be an unfortunate state of the world, of no more significance to me personally than if the demon's largesse had instead depended on my date of birth. (In either case, the targeted state of affairs happens to feature me in it, but not in any way that I am answerable for.)
* Ought implies Can. Here we may draw on the previous two points. It is precisely because we can respond to reasons for action that we are liable for our failure to do so. But even if we have an excuse that exempts us from blame for our agential failure, the fact that we really "ought" to have done otherwise in a sense that implies an agential failure on our part nonetheless retains a special normative significance for us as agents -- a significance that sets this case apart from the sort of 'mere disutility' discussed previously.
Michael Smith's global consequentialism rests on the idea that, just as we can analyse the 'right act' as the best act of those available, so for any other category of evaluand X, we can understand the 'right X' as the best X of those available. But there's an important difference here.
Note that in case of action, there's a naturally privileged class of "available options" from which the 'right' action is selected. (It's hard to specify this modality exactly, but I take it to be something to do with one's rational capacities, e.g. the class of actions that one could perform were one to form an intention to so act. Compare Doug Portmore's "securitism".) The relevant reference class is speaker-invariant, as it is fixed simply by the situation and capacities of the agent in question. By comparison, there is no such natural reference class for assessing mere evaluands such as eye colours. One speaker may be interested in the best eye colour amongst those genetically possible for a child of one's parents, whereas another may be concerned to evaluate amongst all of those humanly possible. There's no speaker-invariant (and non-parametric) normative property out there in the world corresponding to the globalist's "ought". There is such a property for the ought of action. So the parity thesis is false.
[For more detail, see section III of my paper 'Fittingness: the sole normative primitive'.]