While I can agree with your general idea, just noting that I don't think most people probably mean "there's not a single case where torture works" when they say it doesn't. Accuracy and all, yes, but you are setting the accuracy with the question.
I guess it depends on how cynical you are but that there are situations where it works with say 99% probability is pretty shrugworthy. Though experiments excluding a lot of factors and possibilities are great to clarify positions but you are always running the risk of excluding the factors that are important for the real life application.
Or in other words, if someone says torture doesn't work and in 90% of all the cases it doesn't, he would be right. You'd possibly still be able to select 1% of the cases so that in each there is a 90% chance that it works. *shrugs*