One note, Morenk, is that the fitter does very little clock skew manipulation. (There is an option called Enable Beneficial Clock Skew that does this, but only enables a single dedicated skew as the clock enters the LAB, which sometimes can't be done due to other registers in the LAB that can't handle it). A better way to think of it is that the clock trees are laid out first, with an effort to get the least amount of skew. That's the main goal for the global clock trees. With minimal skew, the clock period dictates how long the data path can be, and the fitter does a TON of work to try and shorten the data path. So the emphasis is on the data path, not the clock skew.