The concordant and discordant percentages are just components in the computation of the area under the ROC curve (AUC - labeled as the "c" statistic in the the "Association of Predicted Probabilities and Observed Responses" table from PROC LOGISTIC), which is the most popular fit assessment statistic which ranges from 0.5 for a model no better than chance to 1 for a perfect model. However, like R-square in ordinary regression models, what is considered "good" will depend on what you are modeling and your purposes. So, there can be no hard rule that says, for all possible models and purposes, that you should have an AUC of x in order to have a "good" model.
