These are some graphs I created from the data publically available at CiviCRM's Sourceforge. Sourceforge provides limited data only on the release dates of a version (i.e. 3.1.5) and then the number of subsequent downloads to date but we can still interpret some useful conclusions from the data. Dave Greenberg shared one of these graphs at CiviCon. The data is from Mar 1, 2011 so I wanted to publish them before the data became too stale.
The first graphic is the simplest to understand - CiviCRM all versions 2.x compared to all versions 3.x. Note that the timeframe measured in this graph is similar. As such, we can conclude that the number of downloads for 3.x have increased about 27% compared to 2.x.
The next graph shows the downloads of all versions and their release dates. The distribution here is roughly equal, but remember there have been four minor versions of CiviCRM 3 (3.0, 3.1, 3.2, 3.3) so the 3.x version has more overall downloads.
The final graph I think is the most interesting, which focuses on revisions which are the third number in the series such as: 3.1.5 or 3.2.3. This graph shows a tendency for the CiviCRM community to download more of the x.x.2 and x.x.3 revisions than the x.x.0 or x.x.1 revisions. From speaking with other members of the community, my conclusion was that this behavior is due (at least in part) to concerns about stability of the earlier (x.x.0 or x.x.1) revisions.
CiviCRM Core Team and community are working hard to create automated unit tests for the software that will increase stability of revisions.
My suggestion and approach (which I encourage others to consider) is to also do increased testing for my live client data by setting up test servers using real sites and testing beta versions to make sure features important to my client work. I've already done this for a couple clients and will do more this week.
If you haven't setup testing servers for some of your clients, consider doing so now. You will not only help your client but CiviCRM as a community.