The job of the patent office is impossible in the field of software patents.

It is not realistic to expect anyone to judge novelty and non-obviousness in a field this wide and active. Imagine trying to evaluate mechanical engineering patent applications if millions of people carried machine shops around in their backpacks.

The only solution is to end software patents. Obama could do this today by directing the patent office to respect the Supreme Court's Flook decision, which they reiterated as good precedent in Diehr and in Bilski.

The non-obviousness part absolutely requires more domain expertise than I expect a lot of patent clerks bring, but checking for prior art seems like something that could be found via search or crowdsourcing.

Imagine if there were a public review step to granting patents, where the broader community could provide examples of prior art and debate the non-obviousness of it.

There are lot of additional problems though... the way the Patent Office washes their hands and leaves it up to the courts to sort things out etc. I personally think ALL software patents should be abolished completely. In absence of that, NPE's "non-practicing entities" should not be allowed to hold patents. You need to prove that you are using it (in commerce) to keep the patent. And you should not be able to sell a patent unless it is sold with an existing revenue producing business. The US system has turned into a joke.

Can you name an example of a good software patent?

Mpeg,h264 series of patents. I'm not sure any one would invest the same effort in these particular domains if they knew every one can copy them (though current crop of internet companies like google have an incentive to do it).

As long as it meets the non-obvious and no-prior-art tests, I don't see a problem with the concept.

Something like mouse gestures feel like a patentable innovation worthy of a temporary government-enforced monopoly.

Sure, but not for a decade or more. The timescale of patents is simply incompatible with the timescale of software.

I totally agree on that front. Five years seems like a reasonable upper limit to how long software patents should last.