Multiple Mediation Model using INDIRECT Macro Vs. Amos - different results!

#1
Hi,
I have a multiple mediation model with 1 IV, 6 mediators, and 1 DV (all are continuous variables). I also have one covariate to control for.
I ran a mediation analysis using those variables (plus the covariate) using INDIRECT macro by Preacher & Hayes. Two of my partial mediations were significant in terms of indirect effects (bootstrapping), so that seemed straightforward.

However, I created a model in SPSS Amos that represents the mediation model just described for the sake of model trimming, but added covariance relationships (double pointed arrows) between error terms of all mediators in the model for theoretical reasons (I theorized that the mediators would be correlated with each other for theoretical reasons). All other relationships were the same. My intention was not to re-test individual mediations, otherwise I would have created multiple models to test each mediator, but just to test the overall model fit with the mediators in it (and because my committee required me to use a path model for my dissertation!).

My results were confusing, because one of my mediators that was not significant using INDIRECT now had a significant pathway to my DV (significant regression coefficient), which makes it look like a partial mediator in addition to the other two as found in my INDIRECT output. All other relationships were consistent with INDIRECT results in terms of significance.

As a side note, I used EM to impute for missing data prior to testing, so sample size did not change between INDIRECT and Amos.

How might I interpret and explain this discrepancy between my INDIRECT model and my AMOS model?

Thanks in advance! :)