Jeff, I know BMI is only a very rough indicator and not very accurate due to individual body type variations but I found this idea quite interesting. Wonder if you have seen the info and what you think?
Walter Willett of Harvard (author of "Eat Drink and be Healthy") rejectes the low end of the BMI charts based on all data, claiming that smokers contaminate the low end. He says that after pulling out the smokers from the Nurses Health Study they found the lowest safe BMI to be 17, that is people with BMIs of 17 (nonsmokers) died at slightly lower rates than those with BMIs of 21 to 25.
Check out the discussion on "Optimal BMI" as I cover some of this including the Nurses Health data but yes, smokers contaminate the lower end as do some cancer patients and other terminally ill patients. I would not go as far to say that 17 is optimal but clearly, the lower end of 18.5 to 22 is.