There's a recipe for creating counterexamples to 'ought' implies 'can' in the epistemic case which helps convince me of this. Take your favourite case of someone holding an irrational belief that she epistemically ought not to hold. (E.g. Carrie's believing at 5.40pm that aliens are invading the Earth despite there being absolutely no evidence to that effect.) Then make it so that the subject's holding that belief is beyond her control in your favourite sense. (E.g. Specify that Daniel has attached a brain-manipulating device to Carrie's head so that when he presses his remote control button at 5.40pm she will start believing that aliens are invading the Earth, despite having absolutely no evidence to that effect.) These will be cases where the subject epistemically ought to refrain from believing the proposition in question yet it's not the case that she can refrain from believing it.
I think it's actually rather plausible that there are epistemic 'oughts' that don't imply robust 'cans'; but I don't think the above recipe will yield any good arguments for it. My reason for thinking this is that we never identify an epistemic 'ought' independently of circumstances, but this is, effectively, what the recipe requires us to do -- identify an epistemic ought and then change the circumstances. This seems to me to be a sort of philosophical sleight-of-hand. I also worry about question-begging: The key question is whether the recipe generates coherent cases; but, given the way it is set up, it can only do so if ought and can are detachable in such a way that ought does not imply can. If there were some other way to show that the cases in question were coherent -- that changing circumstances does not change 'oughts' -- then it wouldn't be question-begging; but I don't see what such an other way could be.
UPDATE: In a useful clarificatory comment to a commenter making a similar objection, Jenkins responds by suggesting that epistemic oughts aren't, in fact, very context-sensitive, as can be seen by the fact that we hold stupid people to the same epistemic standards as (for instance) clever people who aren't thinking as they should. But I'm not convinced this is really true; we hold stupid people to the same standards only insofar as we think the difference between them and clever people is not enough to excuse. But if we really look at the sliding scale of IQ, I think it will be found that no one in their right mind holds people with an IQ of 60 to the same standards as people with an IQ of 130 except in the bare basics that they have in common.. In other words, we may hold A, who is a little slower than B, to the same standards as B; but no one can reasonably do so if the difference is drastic. And the difference created by the recipe is, in fact, drastic -- far more drastic, in fact, than the difference between someone with an IQ of 60 and someone with an IQ of 130.