Variable of importance and Q2Y in PLSR

In Partial Least Squares Regression, we can set a threshold to variable of importance scores to extract variables that have significant influence over the output. We can then reduce the model size to include only the variables that are significant according to the Variable of Importance score threshold. In principle, I would imagine that the Q2Y (total variance predicted by Y) of the smaller model to increase, but is it possible to have this Q2Y decrease? If so, what does this mean?