It's also interesting to mention that there are multiple sub-topics related:
1. schema vs no-schema
2. type checking
3. referential integrity
4. data integrity
Now in reverse order:
4. every database should guarantee data integrity (i.e. you get back what you put in).
3. referential integrity is an interesting and challenging problem in the space of distributed systems that has impact on the trade-offs that the system has to make
2. type checking: while I'm not aware of any non-relational database supporting this, it is a feature that could be added at some point by basically tagging specific fields/attributes with typing information. For the time being I think people that use non-relational databases and requiring type checking are implementing it in the database access layer. But there's no good reason why it couldn't or shouldn't be supported by the database itself.
1. schema vs no-schema: IMO these two represent the two ends of a spectrum where what I think would actually make sense is support for optional schema definitions--in a way this could be similar to the way Protocol Buffers/Thrift/Avro define their protocol.
alex @ rethinkdb