One key disincentive here is that building systems takes a lot more work. Academic Human-computer interaction has found itself in a cultural equilibrium in which the “least publishable unit” is quite small. This means that most profs have their names on quite a few papers each year. If you spend 18-24 months focused on a system, your CV may “fall behind” your peers.
Another key issue, which appears to be getting somewhat better, is that Academic HCI over-emphasizes evaluation. Reviewers would often inappropriately insist systems be grounded in quantitative empiricism.
Landay, J. (2009, November 7). DUB For the Future: I give up on CHI/UIST. DUB For the Future. http://dubfuture.blogspot.com/2009/11/i-give-up-on-chiuist.html