Ideally we want an egg that generates as realistic models of realworld  sites as possible, for testing against. Whilst keeping them  very  consistent per profile.                                                                                                                                     
The starting point is to have an egg that installs a set of two profiles, one that models intranet site and the other a public site.                                                                                

https://svn.plone.org/svn/collective/collective.contentgenerator/trunk/collective.contentgenerator

... which as of Jan 8th 2009 has been released to the cheeseshop

http://pypi.python.org/pypi/collective.contentgenerator

... and on plone