A recent Scientific American article (you need to be a subscriber to read it though) demonstrated one of my favorite caveats about inferring whether something is good or bad for you by giving huge doses of it to lab rats (e.g. saccharin, which allegedly causes cancer, and tannins, which allegedly are good for you). As it turns out, hydrogen sulfide, nitrogen oxide, and carbon monoxide, which are all normally considered poisonous, play a vital role in regulating dilation in blood vessels, among other things.
The human body contains (and needs) trace amounts of gold and other things that in large quantities would kill us. What people are doing when they claim something is bad or good based on large amounts of a substance is assuming that the effect is linear (or at least monotonic). However, it seems that lots of things are bad in large amounts but good or necessary in small amounts. Something could also be bad (or neutral) in small amounts but good in larger doses, although it’s not as easy to think of an example. The key point is that it’s important to test comparable dosages.