if I understand correctly the key benefit would be that models could be trained on smaller datasets and therefore reduce the time spent computing the models? I

if I understand correctly the key benefit would be that models could be trained on smaller datasets and therefore reduce the time spent computing the models? I am not convinced that this time saving is more than the time spent to engineer the combined and synthesised data.