Example datasets where PCA could improve or decrease performance of SVM?

Question

Going through the top answers in How can top principal components retain the predictive power on a dependent variable (or even lead to better predictions)?, I understand that doing PCA and keeping the top k components may or may not improve classification performance of some supervised learning algorithm. For decision trees I can visualize how this could be the case, but I can't visualize a dataset (real or synthetic) where keeping the top $k$ components after PCA improves or decreases classification performance in SVM.