1) I find that the outliers make a big difference in any distribution. For example: the vast majority of start ups fail, but the ones that succeed can make a large impact. To your example, the vast majority of compounds tried as drugs fail but the rare ones like peniclilin change medicine as we know it.
2) Personally I've found that people's current status in life can be a predictor of future success but it far from perfect. I would suggest that rather than trying to predict who is going to succeed beforehand we come up with a system that gives them some opportunities to be successful.
What if we really did give people the option of periodic sabbaticals where they could do whatever they want (including "dancing around fires and changing")? It could massively reduce stress levels, and given stress' negative impact on health, this really could help (on average).
If I say dancing around a fire and chanting, on average, doesn't help people with kidney disease and you retort with 'it helped my grandmother' - where do we go from here?
Should I get to individually know every person with kidney disease to see if dancing around the fire will cure them?
That seems to be what you're suggesting, unless I'm misunderstanding you.