I think these are all good ideas. I would focus on building something simple that showcases Cassandra's strengths. I have found that Cassandra is powerful when storing and query large, dense, interconnected datasets. Think about what data is out there that could be analyzed at a deeper, more domain-specific level. (domain being per user, per host, per term, per property of an object).
Facebook's inbox search is a search index per user. Twitter timelines are extremely customized and personalized based on the relationships between users. Digg first used Cassandra for the personalized "green badges" feature by creating a per-user index that stored each story you dugg and which of your friends also dugg each of those stories. Cassandra is good at storing large amounts of data that is a few layers deep.
I think the demo app should not necessarily be feature-rich, but it should show us a slice of data we've never seen before. Maybe it could perform a locale-sensitive search or indexing task. One of the main strenghts of Cassandra is that it allows you to go from an extremely broad, large set of data to a very narrow, important set of data very quickly. Try to think of large sources of data. Chances are you might have to build some sort of crawler/indexer to get this data into Cassandra.
A dumb example could be something like this: Let's crawl the pages 10,000 websites and build an index of page properties (number of images on the page, title of the page, outbound links on the page). Then we could build a interface that let's users quickly drill down through all that data to find the pages that have no images, "dogs" in the title, and at least 5 outbound links. Now, this is not really a useful application in itself, but if you understand that Cassandra lets you drill down from a LOT of data to a very local-specific set of data very quickly, you should be able to think of something truly simple yet useful.
Another example could be to build a service that let's websites pass a user id and page id every time a user views a page on their site. You could then serve back to that site an Amazon-style "users who viewed this page also viewed this page" module. You would maintain a per-page index in Cassandra that, given a page, shows what the most other popular pages are. For a decent-sized website, this is a lot of write traffic that people are having a hard time doing anything useful with in MySQL due to the sheer size of the data (ie pageviews).
Replacing the backend of a blog engine might be something that Cassandra can do, but it doesn't really showcase why people would ever use Cassandra and how Cassandra is good at querying specific data out of large, broad datasets. Think of relationships between objects and properties, that's where the real value can come.
Like I said, the demo app doesn't need to have a ton of features, but it needs to showcase the capacity for handling large volumes of data.