Work out the numbers and you'll see why it isn't. The variance in health care payouts over a pool of N people is proportional to sqrt(N), whereas the mean is proportional to N (see the central limit theorem).
For a family of N=1 to N=4, sqrt(N) is of the same order of magnitude as N. This is bad, since you can't reasonably predict how much you need to pay. This is why insurance makes sense for me (N=1) or a family of N=4.
Once N=10,000, sqrt(N) = 100, so the variance is only 1/100 of the mean. At N=100,000, it's only 3/1000 of the mean. Every factor of 10 increase in the insured pool lowers the standard deviation by a factor of 0.316=1/sqrt(10).