Very interesting thread.
There is a term called 'statistical power' which calculates the minimum sample size required so that one can be reasonably likely to detect an effect of a given size. For nutritional studies, this often approaches 1000+ participants. Now, times this by four (each blood type) and add controls. Provide each subject with engineered meals and pay for admin followup and experimental work.
Turns out some ideas are bigger than the current experimental model.
Yes, there are four blood types, but a study of just one Blood Type (O for instance) would be a step in the right direction.
Select N type Os. For three months N/2 subjects would follow BTD; the other N/2 would follow a lacto-ovo-vegetarian diet (no meat, fish or poultry, but dairy & eggs allowed.)
Test before and after three months:
HDL / LDL
plus a brief self-assessment survey
Let N=40. So what? The two groups are getting nearly the exact opposite diet from what they need, so you'd expect a measurable effect. A group of grad students at BU could conduct this kind study in two successive semesters.
In another year, do Type A, BTD vs Atkins.
Then critics can attack the study, not the diet.
If this were started this five years ago, we'd have some interesting data by now.
Double-blind of course is impossible, but that doesn't mean it shouldn't be studied.
BTW, in industry we typically use N=5 for each group.