The paper you want to have a look at is Thaddeus, GIT and Flips, http://www.jstor.org/pss/2152810GIT and Flips (JSTOR), or Dolgachev-Hu, Variation of GIT quotientsDolgachev–Hu, http://www.springerlink.com/content/3655664711102642/Variation of GIT quotients (Numdam).
There are finitely many polyhedral chambers within the space of possible linearizations. In the interior of each chamber, the sets $X^s$ and $X^{ss}$ are constant. Crossing a wall, these sets change; in nice situations, the two GIT quotient $X^{ss}/G$ on each side of the wall are related by a flip.