The idea is to save enough data that you can have an algorithm that will automatically differentiate between stuff you closed within a couple of seconds and didn't really read and stuff that interested you and you actually read through. And the reason for doing this automatically is it seems very difficult to have the discipline take an action every time you see a page that you may want to save so that you can search through again (such as clicking a box for every such page).
Whatever mechanism it is, it needs to account for stuff read through readers and over multiple computers/devices too. I don't want to miss a search result just because I was reading it on an iPhone or in my Google Reader.
Or instead of caching locally, a search interface that would accept a long list of urls to look through.