If I decided to write a new database from scratch, I would do the same as CouchDB, and pick JSON and REST. Why?

I wouldn’t have to worry about writing drivers and plugging them to different languages and maintaining all the different drivers and their quirks. We take for granted that it just works when we use Oracle or MySQL, but the early years these were painful. Different platforms had different bugs, and some companies (I worked on a such project) paid dearly to write drivers to support their platform.

I don’t have to reinvent things like client-side caching or method semantics, HTTP already does that.

I have existing solutions that deal with caching, proxying, security, load balancing, all that good stuff. I used MySQL for years, still don’t know how to run more than one server. I’ve yet to actually use CouchDB, but I already know how to put a load balancer in front of it!

And of course, I can easily expose it as a storage service, on the cloud or behind the firewall.

So if my client is decoupled from the service, I would opt to use REST, it pays in the long run, even in places it wasn’t done historically (you live, you learn). If my client is tightly coupled from the service, most often I’ll gravitate towards convenience.