Warning: graph- and speculation-heavy post follows.
I don't know what happened, but that trend has been reversed:
After a low of ~1.9 votes/post in March of 2011, the rate has settled into a range between 2.6 and 2.8.
That got me thinking about the network as a whole. For instance, I looked at the perceived drop in quality on Digital Signal Processing and concluded that a steady drop in voting was the culprit. Either good contributors are less motivated to ask, and answer questions or posts with few votes come off as lower quality. (Or both!)
What's a healthy ratio of votes to post? Should a large drop, such as SO saw from 2009 to 2011, concern us? To answer those questions, I queried our database for votes/post, total views, and total posts. I've included upvotes, downvotes, and accept votes because these have the greatest impact on participation. I'm using views as a loose proxy for success. (That's not really fair to smaller topics, such as Poker, but we take that into account when evaluating sites.)
This graph shows votes/post on the vertical axis and the horizontal axis is log scale views. The bubbles represent total number of posts:
The left side of the graph is populated by sites we closed and the right side is dominated by Stack Overflow. Taking out those exceptional sites we get this:
So there's a big cluster of beta sites that get between 2 1/2 and 6 votes per post. On the right side we have some graduated sites with low voting and a handful (Skeptics, Science Fiction & Fantasy, RPG, Programmers, CS Theory, TeX, MathOverflow, Travel, and Mathematica) with v/p north of 5 1/2. There are also a number of graduated sites that aren't extreme in either direction.
Some of the sites on the high side of this metric (Skeptics, MathOverflow, and CS Theory, in particular) suggest that votes per post is a measure of "degree of difficulty" in writing questions and answers on a particular site. On Skeptics, if you don't do a good job of establishing your claim as notable, your question will be closed. If you don't follow the rules on using sources, your answer will be deleted. It's probably our most demanding site. Therefore, many people vote on the site without contributing much in the way of posts.
On Stack Overflow, we see the problem with success. There are far more posts submitted than anyone could hope to keep up with. At best, you might be able to keep up with the questions on a popular tag, such as python...maybe. So we can easily see why so many posts get no votes at all. Anecdotally, that's a problem for people who started contributing later in the game; only exceptional posts gather many votes.
Tentatively, I suggest that different sites will have different "steady state" levels of voting. Drops in voting, such as SO saw between the start of 2009 and 2011, should be concerning for those of us watching particular sites. When this sort of decline is observed, steps should be taken to reverse the trend.
Now, the question remains: what reversed the trend on Stack Overflow? Some candidate changes:
Changes that came a bit late to explain the turnaround: