I haven't the time to go through your interesting PDF, but you might want to consult Andrade et al., "Ambiguous Results in Functional Neuroimaging Data Analysis Due to Covariate Correlation," NeuroImage 10, 483–486 (1999).
This is a general problem in statistics, of course. See e.g. "Effects on Simultaneous Tests of Beta_k," Section 7.6 ("Multicollinearity and its effects") of Neter, Kutner, Nachtsheim, and Wasserman, _Applied Linear Statistical Models_, 4th ed.
|