I've seen this misconception twice on the thread now, so I'm calling it out: IQ is not normally distributed. It has fat tails: the extreme ends of the distribution are much more common than a normal distribution would suggest. Some quick Googling indicated that Terman's data on high-IQ people (1921) showed IQs at 4 SDs (160ish) are about 15x more common than a Gaussian would predict, while at 5+ SDs (175-200ish) they can be up to 1000x more common than a Gaussian.
IQ is actually normally distributed by definition. Actual intelligence, if such a thing exists, may not be. The reason for Terman's finding is that there aren't any IQ tests that are valid for people with extremely high IQ.
Then you get into "What's the definition of IQ?" You could argue that IQ is a theoretical construct that's defined to average 100 with a standard deviation of 15 - but then, if you can't measure it with any tests, and when you do try to measure it the tests come up with different numbers, what's the point?
In my physics courses, my professors were always very careful to stress that "If the theory says one thing and the data says another, it's the theory that needs to change." (Well, unless it's a student lab report that measures the speed of light as different from the commonly accepted value. ;-))
I agree with what you're saying, my point is just that all of the current IQ tests (including the ones Terman used) were not designed to be measures. That is, someone with an IQ of 200 isn't twice as smart as someone with an IQ of 100, which is what the term 'measurement' implies. (C.f. the book Measurement In Psychology, which was recommended by tokenadult a while ago.) Rather, they are designed to compare people relative to one another. In other words, regardless of whether or not there is some underlying thing called IQ, no one (to the best of my knowledge) has ever tried to measure it.