Hi Lucho, I am sorry if I am not addressing or understanding your question. You have a target event rate of 11% which may include include enough cases to develop a good tree classification model without using weights/oversampling. Use the Data Partition node which will automatically stratify by the binary target and given you have 2800 observations perhaps output only training and validation data. Train the tree model and check for stability.
In case of a poor model with no or just a few poor splits consider using the decison processing capability of EM (see the Decision node and also review the predictive modeling chapter - see topics rare events and decision processing which discusses using weights as well)