Hello, your friendly evidence-based medicine researcher here. Probably best if you go read Ben Goldacre's book, as he's a lot more lucid than me, but just to counter some complacency:
look at clinicaltrials.gov
A classic example of Pharma going 'ooh look, we fixed the problem' with some lax regulation.
I don't know what % of trials get registered, but if they do, what was originally published on clinicaltrials.gov (the protocol - saying basically 'this is how we are going to do the trial, thee are the key results we will report') frequently gets cheerfully ignored. I'm writing up a classic example of this at the moment where one of the things they originally said they'd measure (the outcomes or endpoints), didn't really work, and gets merrily brushed under the carpet in the dossier sent to us by the pharma company. Even the peer-reviewed paper its written up in lets them get away with this. Nobody polices any of this, and companies (in my opinion) don't get enough of a kicking for doing it (well, they do in the reports I write :twisted:).
By the time you find a journal willing to run an article that says: "we did a trial, but it didn't show anything interesting because of some complicated assumption about response rates in the stats plan, that later proved to be incorrect" the journal is so small, that the company gets accused of trying to bury the results in some backwater journal.
The journals are admittedly part of the problem, but companies are, in my opinion, just as likely to quietly 'forget about' trials like this. Publication bias is probably the single biggest problem with EBM at the moment - hence that is where the efforts of AllTrials are being devoted.