Do these studies factor in a reasonable estimate for the time value of money? Are their projections of future costs likely to be proved accurate, or do they simply project out exponential growth and assume that there's no ceiling to the percentage of the nation's GDP that can be devoted to health care?

Generally, less healthy people will experience similar complications earlier than their healthier counterparts, or they will experience health consequences that occur far less frequently among the general population. So with healthy individuals you basically get a longer stream of payments and the general wear and tear medical visits followed by expensive end of life treatment. With unhealthy individuals you'll generally end up with a shorter stream of payments and maintenance treatments followed with the expensive end of life care.

It seems like lynchpin of which group would cost more, rests on how future cash flows are discounted and future end of life treatment costs are projected. If the growth of end of life care costs exceeds the discount rate by more than a miniscule amount, unhealthy people save the system money. If it's the reverse then healthy people are less intensive users of the system.