Rethinking Dataset Distillation for Classification: Do Distilled Sets Outperform Coresets?