I've had a suspicion for a while that you could improve a global bounding 'box' if you tried looking for a box, and an ellipsoid that fit the model. You then choose the one with the smallest volume.
With an ellipsoid you can center it nicely. Simply find x^2, y^2, z^2 for each point based on a crude center (eg. that of the box) and you can then minimize
(3V/4pi)^2=(max(x^2-2x xb+xb^2) + max(y^2-2y yb+yb^2) + max(z^2-2z zb+zb^2))^3
according to xb, yb and zb to find a better center.
Not suggesting you should try it, but it's food for thought if you're in the mood
Tim
With an ellipsoid you can center it nicely. Simply find x^2, y^2, z^2 for each point based on a crude center (eg. that of the box) and you can then minimize
(3V/4pi)^2=(max(x^2-2x xb+xb^2) + max(y^2-2y yb+yb^2) + max(z^2-2z zb+zb^2))^3
according to xb, yb and zb to find a better center.
Not suggesting you should try it, but it's food for thought if you're in the mood
Tim