Usability testing is often confused with market research, but the two are very different.
Usability problems can easily be detected from a sample of about five test participants. Very different kinds of people will tend to encounter the same usability problems. More than seven participants is just a waste of time.
Market research issues, things like design-preferences and whether they would actually want to use the product, are completely different. These vary wildly from person-to-person and different personalities/backgrounds. You need big sample sizes to judge them accurately.
Many users perform tasks while performing several unrelated tasks on the same computer at the same time. This may have a strong influence on how they interract with a given product, but it is difficult to simulate.
Similarly, users may normally interact with a product at particular time of the day when they are subject to certain distractions or a certain state-of-mind. These may also have a significant effect on the task but are difficult to simulate.
For example, a participant will be much more careful about a purchase they are genuinely making than about a purchase they are pretending to make for the sake of a usability test.
Usability testing typically last 45 minutes, but many web interactions take place over several days, especially when they involve interacting with an online community