I'm probably going to rot in obscurity down here, since I'm posting so late to the story. However, someone
over here did a really basic analysis with the typical "unemployment by college major" data that the Wall Street Journal put up. They looked at variance in unemployment related to popularity of a major. While the data set was incomplete (they didn't have true sample size, so they used rank, and transformed rank), it showed clear indications that those with the lowest sample size had the highest variance in unemployment. Far from making some broad claims about the utility of a major, it suggests that the less popular majors have big issues with small sample size. A single individual's employment history has far more effect on the statistics of those rare 'terrible majors' than the more populous ones. The only way to make the data trustworthy is to look at it for a much longer slice in time than we typically examine it for.
Also, it's worth putting on your economics hat when you think of modifying incentives like this. The problem with the proposed structural change is it assumes that the government can react to changing incentives faster than an individual can. Where there is demand for labour is a shifting target from year to year, and decade to decade (Hell, it shifts from quarter to quarter in some cases!). By deciding where the incentives are, they government needs to be able to shift them to match need fast enough so when there's a shortage of Psychologists and a surplus of Biologists, people can react to it accordingly. I'm skeptical about a government's ability to react that quickly with policy. If you're going to include incentives, it's best to include incentives for education in general, and not for specific major, so such bias won't occur. If the incentives in the form of subsidization are equal across the board, demand signals should still be seen.
And taking off my stats and economics hats, and putting on my skeptic hat, I want to see percentage-wise how much these 'terrible' majors actually cost the system. My intuition based off of the variance in unemployment vs. rank-popularity is that it doesn't cost the system much at all, and this is much-ado about nothing while the real expenses (Military spending, Medical spending) is ignored. Of course, much of the current fury over debt ignores the fact that the government is not like household/private debt. The two are functionally different.