At SPSS ver 11.0 at the corresponding help file for multivariate linear regression the author insist without referring anywhere that "A variance inflation factor greater than 2 is usually considered problematic, and the smallest VIF in the table is 3.193.". Does anyone knows where such a rule appear? (book, article etc)

2 seems small for the ViF, but it's not my usual metric for collinearity.
I usually use the Condition Index (>10 indicates 'moderate' collinearity, > 30 indicates 'severe').
Rules of thumb are hard to find in the lit, so I can't immediately point you to any. I can tell you that Belsey, Kuh and Welsch "Regression Diagnostics" describe a # of critiques of ViF