I'd say it depends on a number of things:
What format (sensor size), what focal length, what type of lens, and what support equipment do you have.
On 16mm sensor, you're going to get away with a lot more than say S35.
If you had say, a Bartech with a wireless mon setup, your Focus Puller shouldn't have too much issue keeping eyes in focus all the time, though would obviously be harder with still lenses where a slight turn can be 2' of difference. With a wider lens, you'll get much more depth of field. Assuming you're on a 7D, at 7' distance on a 35mm lens set to 1.4, your DOF is 8". 85mm @7', f/1.4, is DOF of 1".
So, keeping wider will help.
Also, I generally find it somewhat imperative to pull off a monitor when using badly marked still lenses for a whole bevy of reasons. In general, I'm not a fan of pulling off a monitor for a whole lot of reasons that don't really pertain to this discussion, but it's somewhat necessary using badly marked (cheaper) still lenses, or lenses that haven't been collimated correctly.
It's going to be easier if you have cinema style lenses on it, as your Focus Puller will actually have real witness markings, and can judge distances, and use marks to focus correctly. With a still lens, not only are the only witness markings ridiculous (like 1ft, 3ft, 5ft, 10ft, and infinity), but there's likr 5 degrees of rotation between each.
Personally, I tend to focus on the eyes as much as possible, and if we can't see the eyes, I'll focus on what the audience would be drawn to watch - unless the DP tells me differently. I'd probably focus on the back of someone's head, simply because it is the closest thing to their eyes that we can see, and the most likely thign for a viewer to focus on. Unless, of course, the DP wanted something else in focus, or if there was a significance of something else in the frame that should be in focus instead.
It's really up to you what you want in focus - what do you think is more natural? What do you think the audience should be watching?