Use varimax rotation.

Orthogonal rotation makes the assumption that there is no common factor. Whether this assumption is defendable depends on the subject matter. If you cannot assume orthogonality or prove it, you should stick with oblique. Multicolinearity is only an issue for very strong correlations. A reduction by PCR should suffice.

Onur,

I would consider two things:

First, PCA assumes that covariance between variables is completely explained by the principal components. In other words, uniquenesses (i.e. measurement errors) are zero. [In contrast, exploratory factor analysis permits non-zero uniquenesses.]

Second, as you note, different types of rotation assume whether or not the components are correlated or not.

So, the question is: Which assumptions would be appropriate?

Correlation does not necessarily implies collinearity; this said, I would (almost always) try an orthogonal rotation first, unless I have strong prior information regarding a possible correlation between the components. This because oblique rotations change some properties of the PCA solution.

You may find help here:

http://stats.stackexchange.com/questions/151653/what-is-the-intuitive-reason-behind-doing-rotations-in-factor-analysis-pca-how/151688#151688

Orthogonally rotated component is the best option. This articles may be of help;

Article Rotation of principal components: Choice of normalization constraints