How do you speed up a queryset that uses .distinct()?

Question

I'm looking to speed up a query that needs to use distinct because it has a M2M field it selects on. At this point I'm not sure if my speed issues are related to how I have my DB server configured or if it's related to my queryset.

My question: What is the fastest queryset and can I also improve the speed by changing my Postgresql settings?

Postgresql Server Information

Instance: EC2 m1.xlarge
Postgresql Version: 9.1
Article Records: 240,695
Total Memory: 14980 MB
shared_buffers: 3617MB
effective_cache_size: 8000MB
work_mem: 40MB
checkpoint_segments: 10
maintenance_work_mem: 64MB

Related Models

class AuthorsModelMixin(models.Model):

    authors = models.ManyToManyField('people.Person', blank=True)
    nonstaff_authors = models.CharField(
        verbose_name='Non-staff authors', max_length=255, blank=True,
        help_text="Used for the name of the author for non-staff members.")
    byline_title = models.CharField(
        max_length=255, blank=True,
        help_text="Often contains an organization. Title of the person, or " \
                  "entity associated with the byline and a specified person " \
                  "(i.e. Associated Press).")

    class Meta:
        abstract = True

class TaxonomyModelMixin(models.Model):

    sections = models.ManyToManyField(Section, blank=True)
    tags = TaggableManager(
        blank=True, help_text='A comma-separated list of tags (i.e. ' \
                              'Outdoors, Election, My Great News Topic).')

    class Meta:
        abstract = True

class PublishModelMixin(models.Model):

    status_choices = (
        ('D', 'Draft'),
        ('P', 'Published'),
        ('T', 'Trash'),
    )

    comment_choices = (
        ('enabled', 'Enabled'),
        ('disabled', 'Disabled'),
    )

    sites = models.ManyToManyField(Site, default=[1])
    status = models.CharField(
        max_length=1, default='P', db_index=True, choices=status_choices,
        help_text='Only published items will appear on the site')
    published = models.DateTimeField(
        default=timezone.now, db_index=True,
        help_text='Select the date you want the content to be published.')
    is_premium = models.BooleanField(
        choices=((True, 'Yes'), (False, 'No')),
        verbose_name='Premium Content', default=True)
    comments = models.CharField(
        max_length=30, default='enabled',
        choices=comment_choices, help_text='Enable or disable comments.')
    created = models.DateTimeField(auto_now_add=True)
    modified = models.DateTimeField(auto_now=True)
    objects = PublishedManager()

    class Meta:
        abstract = True

class Article(AuthorsModelMixin, TaxonomyModelMixin, PublishModelMixin):

    title = models.CharField(max_length=255)
    slug = SlugModelField(max_length=255)
    lead_photo = models.ForeignKey('media.Photo', blank=True, null=True)
    summary = models.TextField(blank=True)
    body = models.TextField()

Querysets I've Tried

Queryset 1

Query time: (76 ms)
Pros: Fast and no chance published articles won't be displayed
Cons: If a higher id has an older pub date then the article list will be out of order

queryset = Article.objects \
                  .published() \
                  .filter(sections__full_slug__startswith=section.full_slug) \
                  .prefetch_related('lead_photo', 'authors') \
                  .order_by('-id') \
                  .distinct('id')

Queryset 2

Query time: (76 ms)
Pros: Articles are in order all the time
Cons: If two articles have the same pub date and time, only one will be listed

queryset = Article.objects \
                  .published() \
                  .filter(sections__full_slug__startswith=section.full_slug) \
                  .prefetch_related('lead_photo', 'authors') \
                  .order_by('-published') \
                  .distinct('published')

Queryset 3

Query time: (1007 ms)
Pros: Articles are in order all the time and no chance of articles not being listed
Cons: Much slower!

queryset = Article.objects \
                  .published() \
                  .filter(sections__full_slug__startswith=section.full_slug) \
                  .prefetch_related('lead_photo', 'authors') \
                  .order_by('-id', '-published') \
                  .distinct('id')

Queryset 4

Query time: (4797.85 ms)
Pros: Not much, however not using DISTINCT ON means it works on other databases like SQLite for tests
Cons: Much slower!!!

queryset = Article.objects \
                  .published() \
                  .filter(sections__full_slug__startswith=section.full_slug) \
                  .prefetch_related('lead_photo', 'authors') \
                  .order_by('-published') \
                  .distinct()