My parents used to tell me that, as working scientists, they far preferred the badly delivered talk replete with lots of excellent data than the reverse. Although, I think we would all agree, the optimal would be a superbly delivered talk filled with brand new, highly significant results that are unexpected within the current paradigm.
Too often, in my experience we have the worst of all worlds: virtually no data delivered in an uninteresting manner. It seems to me this is a matter of training: we should mentor trainees to turn down speaking invitations unless they have some real new data and we should teach them to prepare to deliver talks well (e.g. don’t read your slide deck to the audience).