The resources can be formally defined in relation to one another. Or not.
HTTP has proved to be the most popular URI. It is a protocol for getting stuff, hence the name (Hypertext Transfer Protocol). It makes no assumptions about what you want to get, be it a HTML Web page, PDF, Word document, etc. The most used format it serves on the Web, HTML has "no special place architecturally" within the Web architecture. It is just another format. The format of the data is notated by its MIME type.
The intro also states an interesting, though, thus far in my mind, unconnected factoid: That HTTP runs up against some compatibility problems with distributed object-oriented systems developed by the software development community (As opposed to the Web developers), namely CORBA, DCOM and RMI. Remote Procedure Calls could be a bridge: RDF can work with RPC, namely by considering RPC a structured document.
Taken from a W3C document on design issues for the Web, notably this overview section.