[AMOS] Non-recursive path model with moderator; a little over my head perhaps

Hey guys. Looking in this subforum I see a lot of viewers and no responders, but alas, I must try. My professor can't help me.

I'm helping a student with her dissertation analysis, not realizing how quickly this model got complicated (see attached model). Please note that I have no input on the theory behind this model, simply the statistics portion of it.

1) Because this model is non-recursive (feedback loop) do I need to covary all exogenous variables?

2) In testing the moderator variable for collinearity (first SPSS output on page 2), I find ridiculously high VIF values. I have seen literature for and against mean-centering to help alleviate this problem, but research suggests it really has no effect. Because I can (not the researcher), I removed the moderating variable and all VIF values went WAY down (see SPSS output 2). What is this suggesting?

3) Is this model proper or am I missing major aspects based on its non-recursive nature?

Thank you so much if you take the time to look. It feels horrible having no one in my statistics department well-versed in SEM when they actually teach the class. I'm stuck!