Yeah, it's an important point. There's two primary aspects to head tracking - translational motion and rotational motion.

Tracking rotation can be done pretty cheaply & with low latency these days, and I expect the Daydream viewer will probably handle it quite well.

Translation is quite a bit more challenging, and I'm not expecting much in that department for this iteration. The issue can be addressed to some degree with a neck model that maps rotation to translation, but it can't handle real translation, such as leaning or walking around.

This is on par with Oculus/Samsung's current mobile solution, the Gear VR headset.