[1] I would be shocked if, give two groups of a million, both groups had the exact same number of deaths in a predetermined time period. 1 death difference between two groups that large is pure background noise.
How about this way: you agree that if you drink enough alcohol it will kill you, right? So you could surely find someone who almost drank themselves to death. Perhaps they got liver disease and came very very close to dying, but survived. So there must be a line somewhere. Given a huge enough pool of people, it should be possible to find someone who came so close to death that a single extra drink could have killed them.
([1] It's a thought experiment. The two groups lead otherwise identical lives besides the additional alcohol.)
If the relationship were linear, then you could spread that 500 milliliters out over 100 days and see the same effect, which is plainly ludicrous. The rate of alcohol consumption plays a large role, a single binge can kill you, slow and steady drinking can plausibly extend your life. A single extra drink is also going to do way more harm to a long-term alcoholic than it will to a young healthy Mormon; that Mormon is at absolutely no chance at getting cirrhosis, the risk to the alcoholic is non-zero.
The relationship between alcohol and mortality is something that you could probably fill damn near half a medical library with. "half a liter is a micro-mortality" is silly oversimplified nonsense. It might work for radiation exposure (http://en.wikipedia.org/wiki/Linear_no-threshold_model) but it is nonsense for alcohol.
So long as the line is going up and to the right, it's specific shape only matters if you want to debate the micromort value, not whether the concept exists.
Some of these things plausibly can be modeled as LNT. Radiation possibly can, perhaps airplane rides too. Some of them cannot be, like alcohol consumption.
Its obvious that consumption is positively correlated to certain fatal conditions, but its not at all obvious that the risk is in proportion to (i.e., linearly correlated with) total lifetime consumption.
That's simply not the way things work.
If we had 10e20 people in each group and gave an extra 0.5 liters of alcohol to each one, would you be shocked that a millionth of them died?
Do you really think that a Mormon drinking 3 glasses of wine in his life has a one in a million chance of losing his liver?
If you had two identical groups of a million people each and one group received 0.5 liters of wine per person, would you really be shocked that the group that got the wine had a single extra death?