The Bokode pose estimate seems quite stable indeed, but then the comparison pose estimate (the "G" in the black square) seems to be willfully bad. I've seen better than that, if maybe not quite as stable as the Bokode example, with traditional 2D code matrices.
Unfortunately, the 2-camera rig that they use (one focused at the scene, one at infinity) isn't exactly standard. And it probably won't work with cell phone cameras at all, since these are fixed-focus. Finally, if the camera has to move around a whole lot before it has seen the whole code, it probably isn't going to take off as a barcode replacement. Information at a glance is important in that area.
Besides, I don't really see the need to visually encode a whole lot of information on an object. A small reference ID or an URL is enough, then anything can be looked up online without size restraints.