The main reason for that is that there is an established reputation system for papers, namely journals, but there isn't one for content formats that aren't papers - data-sets, code, videos, images etc.
Part of what Academia.edu is doing is building a new reputation system in science based on audience metrics: how many people have read your papers, from which countries; how many followers you have etc. We are then attaching various kinds of scientific media to that reputation system in order to accelerate sharing.
The idea is that if you can build your reputation by sharing your data-sets, you're likely to consider doing it.
Of course that doesn't prevent you from uploading data to other places on the web, such as the site you mention...
The Research Data Alliance (http://rd-alliance.org/) is trying to solve many of these problems.
My sense is that the most critical aspect here is the incentives aspect. I think if sharing data-sets can enhance a scientist's reputation, and make it easier for them to demonstrate to grant committees that they have had an impact on the field, scientists will put in the work to curate and share data-sets.
Academia.edu started with a lot of promise but is frankly a useless placeholder of your name and Email ID now. ResearchGate has surpassed them long ahead in useful features and Mendeley is already very good. They are just making some money from the academic jobs ads and I frankly don't see a future for them if they are happy being almost the exact same product for the past 3 years I have used them.
Mendeley and ResearchGate are both great companies, but if you look at Quantcast/Compete/Alexa, you'll see that Academia.edu is noticeably larger than them in traffic. Academia.edu also has more users. I'm sorry to hear that Academia.edu doesn't satisfy your use-case though....
Readers prefer papers with data-sets attached to them, because you can check the data and see how the conclusions of the paper stand up. That preference may be part of what drives the higher citation rates.
I heard that http://bohr.launchrock.com/ is working on something similar.