I use Thrift and it is underhyped. I looked at Cassandra, and it offers more than HBase with a lot cleaner and smaller codebase. That's not just NIH, that's some really important thing happening here: standard tools like relational databases do not work for companies like google or facebook. Suddenly to process these volumes of data you do not have to worry about the back-end. That lets small startups compete on the same terms with big guys.

For example JS-Kit.com generates 10G-100G worth of data a day. They had to create an equivalent of Amazon's Dynamo to handle it. Easily half of their resources are dedicated to maintaining and extending this infrastructure. Now imagine if they could just use Cassandra and focus on their core? I do not know what Disqus or FriendFeed are using (probably mysql?), but they can benefit from it too once they grow.

That could affect Google too, there will be more competition from smaller startups like webmynd or cuil because of it. And if startups adopt Facebook's tools instead of Google's, that would make them less attractive to Google as an acquisition target.