I have been reading about modeller usage for awhile and ran some tests with different models from the dual specificity phosphatase family. However, whenever I run the modelling with salign for multiple templates, my GA341 score is many folds lower (under 0.1) and the dope score/molpdf scores are lower as well compared to just a single template. 
The templates that I use all give me a decent 50%+ identity based on output from build profile and I generally use 3 templates for my multiple template modeling.
From reading it seems as if multiple templates should usually generate better scores but running 3 trials of 600 with different sets of templates for different proteins I still get better results from my single template models. So is there something I have to manipulate with the scripts and parameters?
I'm sorry for the vague question because I'm not too sure what kind of additional information to provide but I hope someone can provide me with more insight into this.

Thank you!