Why MongoDB due to centralization of the nodes? Mongo isn't really a fast DB and certainly uses more resources.
MongoDB's speed depends on how you organize your data. If you organize it like a relational database, where every piece of data is in it's own document, then sure, it's not much faster than SQL, and it'll use more resources.
Now, if you store similar data in the same document, it takes much less time to retrieve. Sure, you retrieve more data, but, thats typically less of an issue than the speed at which that data appears.
For example (Simplified):
Say all it took to store the blockchain was a main table (blockchain), and a transaction table. Transactions are parented by blocks, blocks relate to each other by referencing the prior block (this is pretty close to accurate, although there are some other tables that go along with this in most implementations)
In order to pull the transactions of a specific block, you have to query the main table for the particular block ID, then query the transactions table for each transaction that relates to that black. Standard SQL. The entire table of transactions need to be queried.
With a Document Database (like Mongo, or LevelDB), you store the transactions on the block itself, instead of a separate table. In this instance, a query would take much less time, how much less depends on the number of transactions, more transactions, the greater the time difference. Query the block, get the transactions. MongoDB even allows for only pulling back the pieces of the document that you need, so you don't waste bandwidth.
In addition to the speed, you write queries for MongoDB in javascript. So Rise will initially be an entirely javascript stack. We will be supporting other languages for DApp development, but they will be interacting with the DB through Rise APIs, which are all javascript. Any language that can consume a JSON / REST API can be used for Dapp Development, and MongoDB provides the data storage.
Hopefully that makes sense. I'd love to talk to you more about it in our chat if you would like.
https://www.hipchat.com/gtjX3uPYEFor Document DBs, there are a couple of options, but I've been working with MongoDB for a very long time, and I enjoy working with it, it's engineers and support staff, and with it's ORMs in NodeJS. I considered Cassandra, as well as CouchDB, but couldn't find enough differentiators to switch at this point. Cassandra would be good if every Node was going to talk to the same database, since it is designed to scale horizontally, but then that defeats the whole purpose of the blockchain.