In any given test taker how do you measure how much of their score is related to the practice effect?
We can quantify it probabilisticly, such as saying that there is a 95% confidence that, in the absence of practice effects, this individual's IQ score falls between 94-106 (or whatever), given a result of 100, and given no further information about this individual.
I think you're in danger of misrepresenting the operation of p here, presumably because of the limits of posting on a message board. A confidence level of 95% doesn't mean we can ignore practice effects as being only 5% of the total. It means that we can be sure that such results would only arise from random chance, given the statistical assumptions of data and distribution and independent factors inherent in the test being used, 5% of the time.
In order to be confident that practice effects have been controlled for, we have to identify all such effects and understand their impact and the distribution of that impact, in order to correct for them statistically. That's quite a challenge. This is why, in practice, researchers don't tend to do this, and instead rely on presenting a test that is novel.
This is also why, in practice, researchers tend to try and use the test on matched groups where possible. So you try and match group A with all possible confounding factors as closely as possible to group B. But then you hit some problems, because clearly you can't match on a factor that you're actually wanting to test. You can't match on educational attainment, for example, if you want to make a point about correlation between IQ score and educational attainment. Similarly, you can't match on income if you want to compare the effects of IQ on income. IQ researchers have to be very careful in how they construct their field tests, because they face formidable challenges in constructing a meaningful comparison.
Then there is the mechanics of the test itself, IQ tests work by measuring conformity. You take the subject's answers, you compare them to the correct answers determined in advance, and work out the score based on how often one conforms to the other. Or as one critic rather unfairly put it, your score is based on how well your answer matches that of a psychology professor from New York. Now obviously steps are taken to control this. You use a multiple choice format so that personal expression is eliminated. You use questions where there is an unambiguously correct answer. You try and avoid questions of knowledge or questions that assume knowledge, and so on. Nonetheless, if you examine one of these tests carefully, you're going to find at least a few questions where you don't agree with the answer, or at least find the answer ambiguous. This effect is greater the further you are from the assumptions and way of thinking of the person setting the test. So again, IQ researchers have to take a great deal of care to balance out cultural differences so that they can be confident that they are getting a test that's broadly valid across different cultural groups.
But remember what we were discussing earlier? You can't control for what you're trying to measure. So when you look at data such as that presented in The Bell Curve, you're looking at tests that have deliberately not been balanced to produce equivalent results across different ethnic groups, and the book claims the differences that emerge as demonstrating something genetic and inherent about those groups. The problem that plagues studies that purport to show consistent differences between such groups is demonstrating that cultural factors have been eliminated, when the usual methods for doing so, correcting for differences between ethnic groups, are not available.
In short (too late), IQ testing is not a simple or straightforward business, and researchers need to be very, very, careful in interpreting the results. I'm not seeing that level of care in The Bell Curve, which is one reason why I suggest that they go beyond the science, speculating on possible sources for differences, rather than actually demonstrating through scientific study what those sources are.