NLP and content extraction is hard and unreliable except when carefully and continually maintained. RSS is simple for both publishers and clients and allows for simple, decentralized implementations of both roles. Any shift toward heavyweight processing is a shift toward centralization, which is the last thing the web needs right now.