The issue is highly field-dependent. In computer science, it's probably still considered highly unethical, but in my field—chemical engineering, where papers are not normally "published"—presenting the same talk in multiple venues is not normally considered a problem. Other fields lie somewhere in between.
I think that, regardless of the ethical dilemma, the issue is that the more likely you are to have overlap between the two audiences who will hear your talk, the less you want to give the same talk—even if it's ethical to do so. At the very least, you should have enough new material in the later talk that it's worth the audience's time to hear the second version if they've already heard it at the first conference.