If it's not a linier polarizer why does rotating it make any difference?
Rotating changes the angle of the polarizing film allowing you to increase or decrease the effect. The degree of polarization depends on the sun's azimuth and the angle of the camera/lens so you need to control how much filtering takes place. You don't want to over do it because then you can end up with very dark skies, colour shifts and unnatural hues.
Edit: the actual physics is somewhat complex:
http://en.wikipedia.org/wiki/Photon_polarization
As a simplification you can imagine light coming from the sun has polarization in both a vertical and horizontal plane: - and | giving +
When light is reflected off a surface like a glass window or leaves then it is polarized in a single axis, e.g. |
The CPL allows you to control which angle the polarized light can pass through the filter, e.g. if the filter is rotated to this "-" position only light aligned like "-" can enter, vice versa for like in the vertical plane | or angle other angle / \ etc.
If light is reflecting off the window in the vertical plane | then you rotate the angle of the filter to horizontal - and none of the reflected light will pass through, only background light that is also horizontal. Any background light that is vertical will also be blocked, which is why you loose 1 stop of light at least.
If you rotated the filter to be at 45degrees then any vertical or horizontal aligned photons will be clocked with a 50% probability.
A CPL is actually a device that control the quantum mechanical state of photons.
By the way, the same kind of concept is used for quantum cryptography. Imagine you have a box that can hold a single photon, you polarize the photon to be in the vertical/horizontal or one of the diagonal planes. If you have done the polarization then you know what angle filter to use to guarantee the photon can escape the box. However, stranger has no way of finding out because if the photon was say vertically aligned then if they chose a vertical alignment and detected the photon emission then all they can know is that the polarization was not horizontal, because the polarization could be either of the diagonals both with a 50% probability, or vertical. String together 8 or so of these boxes and the odds of someone getting the right combo is remote.
(I got it secondhand along with the lens, which is Tokina)