Two rookie questions about randomforest

Hi,

1. I have experimented with implementing Random Forests with Salford Predictive Modeler, and the R libraries randomForest and caret. Salford Predictive Modeler regularly yields higher r² values than I can produce using the R tools. Does anyone have any idea why?

2. I would expect model r² to go up consistently with the number of predictors used. But I frequently find that the strongest model uses only a few predictors, and r² goes down as I add more. How can this be?

Thanks a lot!