• Prediction markets are currently immature and controversial, and so might have trouble bootstrapping.
• Their legality is problematic. (The IEM had to get a special exemption from the SEC to run.)
• Prediction markets like Intrade currently are structured in ways bad for financial return. (IIRC, the issue is that Intrade offers a very low or no interest rate on deposited funds - the float is a source of profit for it.)
• Long-run prediction markets like many possible scientific or academic questions are not financially viable (see 'opportunity cost'), while sports and gambling bets are inherently short-term, taking no more than a year.
• A succession of short-term markets might help, but then you have the problem that with the natural low prices on 'success' shares, it's hard to make any profit. (eg. imagine a 'cold fusion in 2010' market - it'd be at a penny or two. Suddenly shares double due to a new paper! But because it's so lightly traded, you only made a dime on your prescient long position.)

(Did I miss any?)

Hence, subsidies. Peter McCluskey ran a market-maker bot (OB coverage). Some traders discuss bots; note that they say it's hard to arbitrage Intrade & Betfair in part due to low volume and fees and costs (McCluskey's page mentions that Intrade "agreed not to charge any trading or expiry fees".)