I agree with jakebeal that it matters a lot whether you're doing this for one isolated lecture or as part of a class.
As I see it, the main advantage of this is as a pedagogical tool. It's most useful when you're teaching a class, because then you can give small "quiz" questions as interactive polls, and from the pattern of student responses you can get an idea of whether they're understanding the material. I used this technique in a class I taught and found it quite useful. If most of the class was quickly getting the right answer, I knew I could move ahead speedily; if many students got the answer wrong, I could slow down to go over stuff more slowly.
I think these advantages are much reduced in a single-lecture situation. For one thing, at least in my experience, the proportion of standalone lectures that are pedagogical in nature is fairly small. Second, for pedagogical standalone lectures, you typically have a preplanned talk that you're going to give, with little room to adjust the pace based on audience responses. For a class, you can use results from a poll one day to plan and adjust what you talk about at the next class session, but there's no way to do that if the entire thing is just one lecture. Finally, to get at your main question, I just don't think it's worth the hassle for a single lecture. People have to install an app or go to a website. If technical difficulties prevent them from doing so, you have to either use up precious time from your single lecture, or ignore the problems, which makes the poll results less representative. For a class, it can make sense to spend the time resolving technical issues because the polling mechanism can be used again and again over a period of weeks; for a single lecture, any time spent dealing with that is just cutting into the substance of the talk.
If the talk is not pedagogical, I think there's little point to these polls. You don't really need audience feedback (at least not until after the talk) if you're not trying to teach them something and make sure they understand it as you go along.
I personally used the iClicker device and software, because that was provided by the institution where I was teaching. This is nice because it doesn't require anyone to install anything or look at a screen; they just use the device, which is like a remote control, to respond to polls you show in your slides. Again, though, in a single-lecture setting, it's unlikely to be worth it to pass out iClickers for everyone and then collect them at the end. The only way I could see this being practical is if your talk, although standalone, was one of many such talks at a conference or something, and the conference provided the iClicker or other poll infrastructure. Then the economies of scale could make sense, because people could participate in polls across many talks. For this to be worth it, though, you'd have to have substantial interest from the speakers, or else they might not bother to include polls in their talks.
