Riak is a great product/technology. I have no idea what the future of Riak looks like, but Riak in it's current form is something I still highly recommend. Just keep in mind that Riak works best for pure key/value workloads -- most of Riak's other features are (as they've been for years) hit/miss.
I used Riak for several years and was a big fan (still am of that version) but I migrated to CouchBase because it had some key features that were important to me that Riak lacked (Riak was just a KV store at that point).
I'm very happy with CouchBase and I feel like they are way ahead of everyone else in terms of skating to where there puck is going to be. In fact, at the time I was using Riak they were moving very fast engineering wise and building great stuff- and CouchBase was still figuring out what it wanted to be when it grew up (with the merger of Membase and CouchDB/CouchIO). Now the roles are reversed. CouchBase is the one that's moving fast without breaking stuff.
To be honest, I think CouchBase is the secret sauce of what I'm doing. I laugh at all the MongoDB posts on HN especially given so rarely that CouchBase gets mentioned.
I would also like to know the answer to this. Specifically, riak CS. It seems to have been passed by other technologies in the space. It also has some really bizarre requirements (like 1 disk per node).
Unfortunately, Riak CS (as of today now rebranded Riak S2, apparently), is something I don't have enough personal experience with to really have an opinion on.