Enforcing locality on the web is presently impossible since there is no neighborhooding on the web. Consequently, users often initiate web searches when they don't know exactly what they want and aren't sure exactly what's out there. Since, apart possibly from the search engine companies, no one really knows what's out there, everyone is unnecessarily adding to the bandwidth crunch on the net. It should suffice to state a general area of interest then dynamically evaluate recommendations automatically selected from the ocean of available, but as yet unseen, information. This is impossible in a search-based model of resource discovery unless you first know exactly what to ask for. What you need is a map.
For a map to work well, however, browsers must distinguish between at least two kinds of links: links in the sense of "this page finds that page of interest" and links of the type generated by the linkage mapper (or generated by the user) which essentially say: "this page is semantically related to these other pages according to my user."
This distinction between links is the same as the standard AI distinction between IS-A and HAS-A links: IS-A for hierarchy, and HAS-A for property (example: Tweety IS-A bird, Tweety HAS-A beak). Today's web links are mostly HAS-A links with very few IS-A links. IS-A links typically occur only within a particular website or at the topmost level of commercial search engines. The reason is obvious: those are the only places where human beings are doing the mapping.
Browsers should also distinguish between two kinds of link following. The normal behavior is the now common one of each link leading instantly to anything, anywhere, no matter whether related or not. The new one could be a "rubberband" weblink that would make it harder to move out of a neighborhood the further you wish to jump. Users should find that a handy way to limit the otherwise irresistible urge to jump completely out of the current neighborhood they are in and follow an ever-increasing series of links until they are hopelessly lost in webspace.
Of course, those who like drowning in the world's ocean of information can simply turn off the rubberbanding. Finally, the rubberband effect should adapt depending on how many times that particular user wants to get from one locale to another. The mere fact of continued attempt could make jumping between them easier.