It is much better to have a flaming failure than a so so success.
[...] the expriments should be designed so that a failure is informative and a success leads to a major breakthrough. Doing small incremental experiments with predictable results is much less valuable than taking a big risk that will potentially lead to a much bigger reward.