One of the challenges of advising students is working with students whose "quality control" expectations do not agree with that of the advisor. When such mismatches occur, how do we encourage them to provide "better" quality work. Note that it's not the number of hours being worked the concern, but rather issues like returning a marked-up manuscript with half of the important suggestions left unaddressed, or leaning too much on the advisor or more senior members of the group for help.
I recently heard a nice angle on this problem. Computer Scientist Olivier Danvy, adapting Buddhist philosophy, stated in a recent presentation that you need to inhale before you can exhale. What he meant was that in order for grad students to produce quality work, they need to see, read, ingest and understand quality work. Perhaps you can show them the difference between the high quality work and low quality papers you've also come across. Try to get the student to gauge where his/her work fits in relation to these other papers.
You'll also need to explain that even though poor quality work can actually be published, that this is not a good thing. Not good for science. And not good for building a career.
Many iterations of corrections is important. It takes time. But for every comment that is ignored, you need to find out why it is ignored, and make the student aware that it takes your time and is annoying if you need to make the same comment over and over again. Maybe the student does not understand the comment. Maybe the student is overwhelmed by the vast amount of comments. Maybe that just missed it accidentally. Maybe they just disagree.
In my experience, teaching "quality control" in larger-scale projects is part of the enterprise, since (in mathematics, for example) standard coursework provides no inkling of this. That is, it is not typically the case that sloppy or flawed homework or exams are returned with detailed comments, for iterated corrections, to be repeated until the thing is acceptable. Rather, as we know, schoolwork is presented to students as a high-volume stream of disconnected small tasks, most of which truly do not merit "perfecting", but, rather, treating as a bulk-processing problem.
So the methodology and style of iterative improvement and "perfecting" a larger, months-or-years-long project is arguably a novelty to the student. The "solution" seems to be to just keep iterating the corrections, perhaps making the auxiliary point of the inefficiency of your making the same point several times.
I've had the opposite problem a few times, as well, namely, exaggerated attention to over-perfecting an initial fragment, effectively avoiding addressing the sequel and larger project.
Thus, I think that imparting a functional sense of editing and quality control is part of the task of the mentor/advisor/supervisor, although, yes, energy allocated to this takes away from the more literal scientific/intellectual tasks. Advisor's firm repetition of the standards is essentially the only constructive response, I think, since, for example, it seems infeasible to hope that novices can sufficiently critique each other, as they usually share the same inexperience.
I would cite one of the maxims in industry. Your people is only as good as your structure. A good structure designed to give incentives to current students and guidance to new ones can do wonders for any organization.
You can streamline a lot if you have a whole system in place where you have older students mentoring fresh students (one to one basis) and a more structured way of presenting the job. This way the students (and the adviser) won't be running to catch deadlines and will be way more efficient.