As pure scientific research, no, the show wouldn't count as a credible source of data. While they are obviously careful and attempt to be thorough in their methods, as Mr. Savage says, their data sets are so small as to be statistically useless. Their experiments would count as anecdotal evidence, as best. There are all kinds of potential experimental errors introduced into the show (if nothing else due to time, budget, etc. constraints). You certainly wouldn't (or at least, I wouldn't) cite them in a scientific paper or something.
However, depending on the situation, they could certainly qualify as a credible source of additional information, particularly when taking a skeptical approach to commonly-held misconceptions. When they take a statement such as "its impossible to do X", or "if you do Y, Z always happens", they merely need to demonstrate one example of that statement being false to debunk it. They do so in the presence of a recording device that provides clear and concrete documentation of their process and results, which makes it pretty strong evidence.
In general, I find it much more reasonable to use their successful experiments as confirmation or a fact, than to use their failed experiments as a refutation of a fact. For example, I would be willing to say "yes, it is possible for a tax cab to be blown over by airplane exhaust", because they've documented exactly that happening on camera. I'd be much less willing to say "no, it is not possible that Archimedes was able to set fire to a boat using just mirrors" based solely on the few times they've attempted it. In that case, a data set of "a few failures" is nowhere near strong enough to base a conclusion on.
In general, they are much like any other source of information. If you believe the show, its cast and crew are honest and credible people, then you will accept the evidence they have on tape as credible. If you believe them to be faking things for television, you will not. And even if you do trust them in general, you still need to factor in the context in which you're hoping to use their evidence, and how thorough you believe each individual experiment was in covering all possible conditions.