I have got the following dilemma. When I used traditional PLS I got AVE (average variance explained) of 0.62 for a construct, but when I used consistent PLS I get 0.46 for the same construct.
Why don't you try and use AMOS. If you can share the dataset with someone you confide in ..may be you can check the same with a different software to check the reliability issue.
if you have nonlinear relationship or quadratic relations is better to use warp pls, other wise try to use smart pls. its important to know what type of correlation you have.