The duration of the patent in my terminology is also an implementation issue and not a conceptual issue. I agree that in today's world, twenty years are a bit too long. Or rather inventions that can be so readily made should be failed on "non-obviousness" basis again.
The real issue is this: Not the patent system, but the companies are interested in protecting "everything" they do. The law does not provide protection for everything. Copyright applies to specific material or implementations while patents apply when those three constraints are met. There are clearly things that are not protected by either and they should remain as such. The issue is that they don't (and both companies and patents office are responsible for this).
To fix (the implementation), the bar on non-obviousness, novelty, etc. as well as the duration of the patent needs to be made commensurate with the reality, taking into account the low investment costs and timeline that you mention. Maybe even by having varied durations for different domains just as the non-obviousness bar is different for different domains just by virtue of being unmeasurable.
All in all, I have not heard any real reason why software patents should be distinguished from other patents, other than possibly having the bar on them and duration adjusted accordingly.
Edit: Disclaimer: I am not a lawyer.
There are two issues here:
1. You assume the implementation is fixable, I have seen no sign of it so far, especially for software and method patents. Even more so as the number of patents applications is skyrocketing worldwide and intellectual propriety offices have neither the resources nor the qualifications (at least in software) to do a good job at it. This is not helped by
2. Patents themselves have little value if there is low risk and investment in what is being patented. That's the case for software (relatively). Even more so because reinventions are very common in software, if you have a given problem there will be converging solutions independently devised. Patents were originally thought of to grant a short-term monopoly to a creator in exchange for that knowledge being public. The goal was twofolds, to encourage potentially costly innovations and to ensure that knowledge gets into public domain at one point, to benefit all.
I submit that not only is the implementation broken, but the concept itself is broken for any domain with low investment costs and high innovation rates: for patents to "work" their duration would have to be so low (months, or even weeks) as to make them not worth the time.