I've been testing 3D tracking on people on the street (May day celebration) and once the operator understands what it is doing, it works shockingly well as long as there is no obstruction in the line of sight between the subject and the camera. If the subject is temporarily occluded, the camera tends to lose the subject and continue on something else in the immediate surroundings. Restarting tracking after the subject is again visible is virtually instantaneous on the D5 and 70-200/2.8E. I think for this type of subject matter the resulting focus accuracy is actually better than I can do with manual control of focus point. This is because the system uses all 153 focus points and is able to compensate for small lateral subject position changes better than I can. However it cannot handle blocked view of the subject or a pirouette in figure skating so for that, auto area AF works better. Auto area AF isn't quite as precise as 3D tracking when the conditions for successful tracking are in place though.
I've had the camera for almost a year now and I keep finding features of value that I had not thought to use because they didn't impress in previous generations. Tagging a subject and tracking it with recomposing and zooming as the situation evolves is something quite amazing. What remains for the photographer is to know the limitations of each feature. E.g with FX one needs to be aware of what the camera is likely to do when the subject's face leaves the AF sensor array, for example, and how to predict moments where there is likely a discontinuity in tracking and how to recover from it (it is easy). And when not to use 3D tracking.
What is interesting is that Nikon chose to show the active point as one of the 51 manually selectable points instead of showing the in-between points when they are being used. The result is that sometimes the point that is shown as active is outside of the subject but the subject is still precisely in focus. The reason is that an in-between point was selected by 3D tracking but the camera or viewnx-i show only the nearest point in the 51 instead of one of 153. I think probably the correct point is stored in the exif but for some reason it is not shown. Maybe a future firmware upgrade and software update will show it.