I am slightly unimpressed with A9's Opensearch. A standard protocol for publishing search interfaces is a good idea. Whether basing it on RSS 2.0 is a good idea remains to be seen - but at least something somebody calls RSS is widely deployed and it is also extensible so that a search engine may extend the metadata published in search results.
What is decidedly underwhelming is the Opensearch aggregator A9. Seeing the data overkill of a 5-way multi-search aggregated into an A9 user profile brings me back to pre-google days when everybody, mistakenly, thought the problem was finding the right data. That's not the problem. It's not finding all the junk that's hard. A9 is like a browser based version of those desktop super searchers that were popular back in the 90s. And like those tools it is quite simply solving the wrong problem.
The next question then is what the right problem for Opensearch is if it is not the Opensearch aggregator. Personally I think the Opensearch search profiles will be extended with some kind of search profile indicating the grammar of searchable assertions (e.g. a specification telling me that I can search a particular database for the address of post offices based on postal codes). My search for post offices will then lead to this search profile and I will be able to use it. It will be sort of a weakly linked version of the semantic web. The inly version of the semantic web that could ever work.