I personally think it is overstated that most of our content comes from short-term contributors. What I do find is that short-term contributors begin a lot of our articles, yes, but they generally produce content that is not up to a high standard. My experience is that in most fields of interest, large numbers of articles are substantially written by the same small subset of users.
On the other hand, these people have to start somewhere. I think that people get into Wikipedia in one of two ways. The first is that they hear about Wikipedia as a project and go there because it sounds interesting. The second is when they find an article that's not very good, and improve or replace it.
In the first case, people are likely to read featured articles and other stuff pointed-to off the main page first. In the second case, they are less likely to - although they are likely to poke around similar content till they find a 'good example' to work from. In the first case, the FA process is very valuable - we're showing people examples of what to do to create a good Wikipedia article. In the second case, applicable WikiProjects are good, because they improve the chances of someone finding a similar article to a bad one that's done properly.
Articles don't develop linearly - I agree. However, what generally happens is that drive-by users add snippets over time until someone wanders by, thinks "What a mess!" and decides to yank it out by the roots and start over. However, that 'start over' will generally contain most of what's worthwhile about the previous version - people are averse to dropping content (even when it should be dropped). As the article gets longer, these rewrites tend to be of sections rather than the whole article.
-Matt