In the last couple of weeks I've been finally building my product as a web application mainly because of these factors: currently I have more experience on web development that desktop development; I use a lot of SVGs and tables and they are easier to do on HTML; I have my source code in Python and C and it is a pain to distribute Python; cross-platform desktop development tools are not great nowadays (Qt is the only serious option and it sucks); I would have to distribute lots of libraries with my application.
The way I see things now is, if you really want to test the market for something and you can use a web app, do it. It is easier for you to control the updates, iterate faster to improve your idea, and have better discoverability. It is less painful for a prospective client to click on your URL than thinking on downloading and installing your application. The funnel for web applications is smaller than for desktop apps [1], and that is really important when you want to test your application.
Later in the game, after you have a stable idea and enough loyal clients, you can always move to the desktop or mobile if that's the case. This is what is happening with all major SaaS nowadays, they start on the web and eventually go to the desktop when they reach some maturity (Evernote, Slack, etc...)
[1] http://www.kalzumeus.com/2009/09/05/desktop-aps-versus-web-a...