great post. direct and to the point, although there are many more flaws that I am sure you could have shared.

we tried MongoDB to consume and analyze market feeds, and it failed miserably. I can add a couple of things to your list:

* if there is a pending write due to an fsync lock, all reads are blocked: https://jira.mongodb.org/browse/SERVER-4243

* data loss + 10gen's white lies: https://jira.mongodb.org/browse/SERVER-3367?focusedCommentId...

* _re_ sharding is hard. shard key is should be chosen once and for all => that alone kills the schemaless advantage

* moving chunks between shards [manually or auto] can take hours / days depending on the dataset (but we talking big data, right?)

* aggregate (if any complex: e.g. not SUM, COUNT, MIN, MAX) over several gigs of data takes hours (many minutes at best). Not everything can be incremental..

Those are just several. MongoDB has an excellent marketing => Meghan Gill is great at what she does. But besides that, the tech is not quite there (yet?).

Nice going with Riak + PostgreSQL. I would also give Redis a try for things that you keep in memory => set theory ftw! :)