>>Can you compare the actual execution plans with and without group by to find out what takes that long?
>
>I can only do it without the group as the other one never finishes. Without the group, I do not see anything wrong except for the fact that it tries to use some combo index that someone wanted to insert in the tables recently. Would there be a way to tell SQL Server to avoid using certain indexes on such query?

I just re-read your original question again. You don't need to group by ChildPK as in your example (unless it was a complete mock) you seemed to take the first record per Parent Key, not by Parent and ChildKey combined.

If so, you need to use my suggestion.

As for your question, I think it's possible, but I need a quick research. In the worse case scenario you may want to use optimizer hints.