I've tried MongoDB with the C++ API. I had a lot of problems in particular with the BSON objects and then switched to Memcached (+ MySQL). Do you think that the "production" version is working ok for all API's?
Sluggy.com has been using MongoDB in production (with Python as it's main API interface) since ~July '09.
We haven't had a single problem with Mongo or API interfaces. This replaced MySQL + Memcache and was a MASSIVE difference in system load and performance.
Your mileage may vary, but it's very, very rare that I can say I haven't had a single real problem with a third party package (especially one as complicated as a database-type system), more so with early adoption.
(I'm still using Memcache for a few legacy pieces, but in general things I was having Beaker stuff in Memcache such as sessions and Pylons template caches are going into the Mongo Beaker plugin I wrote - http://bitbucket.org/bwmcadams/mongodb_beaker/ )
The C++ API should work well as it is what we use in the database. The only issues is that it depends on boost, and any binary copy of the library depends on a specific version of boost. You may want to try compiling libmongoclient for your computer. If you still have issues, please post to our user list or ping us on irc.