How do they bias the results in that case?
Merck-funded researchers trivialised the problems of Vioxx. Those researchers were not evil or stupid. Subtle influences are hard to spot in yourself.
When you look at the data there's a remarkable gap between independently funded studies and company funded studies. It's very clear that studies funded by companies more often show benefits and less often show harm. (There are several published studies on this.)
And even if the blinding and randomising is effective there's no requirement to publish. Thus, a company can run ten trials and only publish the one that's positive.
Some explanation and citations here, along with many other statistical tricks:
http://www.refsmmat.com/articles/statistics.html#truth-infla...
There's also a well-known problem called "vibration of effects" where you can play with controlling different variables, using different analyses methods, and so on, resulting in an estimate of a drug's effects that oscillates by a factor of two. Look for John Ioannidis' paper "Why Most Discovered True Associations Are Inflated".