There is a good article about it here:
http://www.cptech.org/ip/health/econ/dimasi2003.pdfThey peg the cost of bringing a new drug to market at about $800m on average. It's a roughly 40%/60% split between pre-clinical and clinical phases. So even without any testing at all you're still looking at several hundred million dollars of expenditures. And testing at a lower standard than the FDA currently requires can only save you so much. The paper shows that about 50% of the clinical costs are just from the cost of capital (having expensive capital lying around while waiting to go through testing). It's hard to imagine any practical sort of testing that doesn't incur these costs, even if you deviate from FDA standards and involve less patients, etc.
Also, you have to look at the cost avoided through rigorous testing. Human lives are expensive, purely from an economic standpoint. Say the average 50 year-old makes about $50k/year and has another 15 working years. That's about $750k in income alone. Since an employer would never hire someone that wasn't worth at least their salary, the economic value of the worker is probably well over $1m.
Now you have a study here about a new cholesterol drug: http://www.npr.org/templates/story/story.php?storyId=6577591. On a test of 15,000 people, with half taking the new drug and half taking Lipitor, the new drug resulted in 60% more deaths, or about 0.41% as a fraction of the overall test group. And this was a drug already deep into human trials.
Now, consider that a popular drug like is prescribed to tens of millions of people. What is 0.41% of 10 million? About 41,000. At $1m a pop, that's $41 billion.
Now, I'm just throwing some numbers around. The point is that even extremely small variations in the effectiveness of drugs can have huge economic impacts when you consider how widely prescribed these drugs are (to a demographic, I might add, that's often at the peak of their earning power). And there is just no way the market would be able to uncover this information without rigorous clinical testing. A 0.4% difference in efficacy can have billions of dollars in impacts, but if you're talking about a sample group where 2-3% might die in a given sample period anyway, nobody would ever know about that difference.