Oversampling Tabular Data with Deep Generative Models: Is it worth the effort?