I have hypothesis that utility maximization is always a second order process - there's always some underlying selection process with its fitness, and only because it promotes traits that make agents agents act in a way that best approximates utility maximizing, adaptation executers seem to us like utility maximizers.

Now let's apply this to political systems:

In most political systems ruling elites could only be removed by either military intervention from the outside, or revolution from inside. That's the fitness function. So traits that would be promoted were those related to having strong and effective military, and skillful diplomacy to avoid external threats; keeping population in conditions bearable enough to prevent a revolt; strong policing and effective divide&conquer approach to keep any revolt from succeeding.
In modern two-party (or multi-party with two big possible coalitions) representative democracy economic performance of the last term can predict voting patterns very well - if people's incomes and well-being are improving fast enough, people will vote for the ruling party. If they're deteriorating, or not improving fast enough, people will vote for the opposition (Bread and Peace Model). This means parties' fitness function is strongly linked to short term economic performance, and therefore policies that improve people's well-being will be selected.
In direct democracy with most important decisions being taken directly by voters, there's no selection process - so they will be as unsuccessful as dictatorships. This is extremely surprising prediction to me, but remarkably bad economic performance of California seems to confirm this.
I guess that in single party regimes, like those of post-Stalin communist countries party leaders would be removed in case of highly unsuccessful performance. This would produce weaker selection than in two-party representative democracies, but stronger selection than in pure dictatorships, or direct democracies.
Libertarian idea of free competition between political systems never existed, so there is no need to discuss it.

There are also some hints how to design better representative democracy:

The closest a representative democracy is to proportional one-person one-vote system, be stronger the signal. So countries with highly distorting electoral systems like US would have weaker selection than countries with straight-forward electoral systems like Israel.
If people vote for reasons not related to improvement in their well-being, like for ethnicity, religion, or ideology, it weakens the signal.
A mix of direct democracy to resolve non-economical issues (like gay marriage, abortion), and representative democracy to actually run the country (and politicians wouldn't need to bother with ideology) might work even better. It seems vaguely like what Ireland is doing, and they also seem to be performing very well economically, is it just a coincidence?

I used to think that direct democracy would be a major improvement relative to what we have now, but this analysis suggests that representative democracy (with small bits of direct democracy thrown in) should work much better.