In general, philosophers take it for granted that a person morally ought to X if and only she can X. I'll treat this as equivalent to claiming that a person can have an all-things-considered moral duty to X if and only she can X. So, e.g, I cannot have an all-things-considered duty to shoot magic fireballs from my fingers because I am physically unable to do so.
Outside of ethics, though, there seem to be a few cases where intuitively we don't accept this, or, at least, I don't.
1. Consider norms of epistemic rationality: Suppose the evidence available to me strongly justifies belief in P–it is epistemically irrational not to assent to P in light of my evidence. If so, then I ought (epistemically) to believe that P. However, it's not clear in exactly what sense I can believe that P. Belief voluntarism is probably false (despite the existence of motivated reasoning), so let's assume here that I cannot just choose to believe that P. So, I cannot believe that P in the sense that I am unable to bring myself to do so, at least not immediately. (And I am rationally obligated, let's say, to believe that P right away, given how excellent my evidence is.)
Still, you might say that I am in some sense physically or psychologically capable of believing that P–it is not incompatible with some important or essential description of my psychology that I believe that P. My brain is capable of putting P in its belief box (though not by immediate voluntary assent). I just don't happen to have P in the box right now and cannot simply put it there by will.
However, suppose that I have some sort of neurological defect such that I become physically incapable of believing that P even in light of the evidence. I am otherwise a normal epistemic agent, but I just can't be made to believe that P, though I have overwhelming evidence that P. My belief box is broken and can't hold P in it anymore. Here, I'm still inclined to say that I epistemically ought to believe that P, though I can't believe that P. However, I'm inclined to add that my lack of belief in P is not obviously blameworthy. I violate norms of epistemic rationality–and the norms really do apply to me–but I am not blameworthy for doing so.
This is all tentative, but these are my starting intuitions with epistemic rationality. I'm a bit worried that it will be hard to maintain "ought implies can" with epistemic rationality without also having to accept some form of belief voluntarism.
2. Grammatical/Syntactical Oughts: There are norms about how to put sentences together. This first sentence is grammatical. Second this isn't sentence.
Suppose Agnes suffers a brain injury that prevents her from being able to form grammatical sentences. She retains an understanding of English semantics, but cannot abide by norms of syntax. So, when she wants to express an idea, she'll say the needed nouns, verbs, adjectives, prepositions, etc., but won't put them in the proper order. (Apparently, there are real cases of people having this problem after brain injury.) It still seems to me that Agnes ought to speak grammatically, though she can't. She violates norms of good grammar–and the norms really do apply to her speech–but she is not blameworthy for doing so.
If there's a difference between these two cases and moral cases, perhaps it's that in general, violations of moral oughts are blameworthy.Like