1) John sees a child struggling in the water but decides not to bother saving the child. The child drowns. Unbeknownst to John, a strong man standing at the shore was watching him. Had John run toward the water to attempt a rescue, the strong man would have tackled him and prevented him from saving the child. As it turned out, the intervention wasn't necessary...
Most people, I think, will say that in case 1 John isn't responsible for not saving the child.
Let's distinguish two questions. First, we can ask whether John is a 'free', morally responsible agent. Was he in control of himself, and hence answerable for his decisions during this episode, or was he temporarily brainwashed, hypnotized, or otherwise under some compulsion that prevents his behaviour during this episode from really being attributable to him as an agent? In other words: was John exercising his agency? Call this the 'agential' question. (I take it the answer to this question is straightforwardly 'yes'.)
Once we've established that the agent freely exercised their agency, we can ask a further question, namely: what subsequent events should be attributed to this exercise of agency? In this case, we're interested in doling out responsibility for the bad outcome of the child drowning. So the 'outcome' question is this: is John to blame for the child's drowning? Here the answer seems to be 'no', since it turns out John couldn't have saved the child even if he'd tried. (Of course, it still reflects poorly on him that he didn't care enough to try, so John is certainly blameworthy, even if not blameworthy for the drowning per se.)
So far, this all seems clear enough. The original question, "Is John responsible for failing to save the child?" strikes me as less clear. But I do not think there is any further question here; rather, it just strikes me as ambiguous between the two more clear-cut questions noted above. (Does that sound right? I'd be curious to hear any alternative interpretations here.)
A more general point: I'm wary of questions that seem to turn on the "guise" under which one describes an action. This just seems like a needlessly messy way of talking. Better to refer to actions de re rather than de dicto: we may speak of an 'exercise of agency', or 'what the agent did', and then it's a perfectly clear and determinate matter whether they were morally responsible "in acting as they did." No need to get into tangles about whether they're responsible for the action under this guise or that.
As previously noted, there seems a lot of unhelpful ambiguity in the traditional treatment of responsibility as a two-place relation between an individual and the acts or outcomes that they are “responsible for”. It seems better understood as a three-place relation between an agent, an action (or period of agency, picked out de re), and the outcomes for which their action renders them liable. This extra structure more clearly separates the 'agential' and 'outcome' aspects of responsibility.
(To illustrate: suppose Jekyll knowingly takes a drug which causes him to go berserk, and subsequently kills a man. We can then say that he is responsible, in taking the drug, for the man's death. This is much clearer than the traditional claim that he's "responsible for killing the man", since there's an important sense in which the killing was not a free act or 'locus of responsibility' at all, done as it was under the influence.)