I've been playing with a Data Explorer query to investigate this question and here is what I've come up with: http://data.stackexchange.com/stackoverflow/s/1522/tags-with-most-valuable-accepted-answers.
This monster query will look at the accepted answers on questions from the past year and calculate the score median for each tag. Originally, I was calculating the average, but it tended to be skewed by single questions with an extremely high score. I also restricted the query to popular tags that got at least 1000 questions in the past year.
Looking at the results, we all need to learn Haskell. The score of a typical accepted answer is 5! Coming from my corner, this is pretty unbelievable.
Scala and Mathematica experts are also pretty well off: score median 4 is pretty nice.
OK score: Perl, Delphi, R, F#, there you can still expect three votes for your answer.
Now coming to the lower end of the scale where the majority of the tags is located. You get only two votes for: C#, Java, C++, Python, SQL, C, and Ruby.
There are quite a few tags where you typically don't get any upvotes at all, but none of them made it into the list due to low question volume.
Edit (2019-07-11): I updated the query to make sure it still completes with the current amount of data: https://data.stackexchange.com/stackoverflow/query/1076022/tags-with-most-valuable-accepted-answers