One thing I usually consider is if DBMS installed and provided in the Cloud being analysed for possible in-productivity use supports triggers, partitioning, complex indexing by use of regular expression models etc.

As for number of joins, records, tables etc. it has more to do with the quality of the database design and normalization level than with the cloud solution that provides the database system.

I too would like to hear of some key-value type DBs in he cloud if someone after me comments and has experience with them.

Cloud databases for example Azure (Microsoft Cloud) table is key value pair model. Some how it is good and based on no schema. From my experiences of Azure based modeling, there is no indexes like relational model. But it has two values (Partition key and Row key) are basics to the query optimization. One way looking it is relation Primary Key is partitionkey and rowkey can be used for finer granulite . I found Azure table performs better than relational model because it doesn't have to traverse the binary tree for key values.