I wonder if a similar architecture could be used for building a distributed database that could rival what was lost with foundation.
The key with transactions is that they make life easy for the client. AP systems are easy for database engineers to write (and yet they still manage to screw them up :P), but systems that support ACID constraints are easy for application developers to use. That's why FoundationDB was so special; they promised the best of both worlds; the horizontal scaling of traditional NoSQL systems, with the ease of use (w.r.t. reasoning about concurrency) of SQL/ACID systems.
An oracle replacement won't be built for financial services companies.
If you want those same features, consider VoltDB. It makes different tradeoffs on what kinds of ops are fast, but has all of those features.
Essentially arbitrary two-key transactions are slower. Partition-friendly writes and global reads are faster. Global reads with real SQL can be orders of magnitude faster.
You probably meant [a-zA-Z]+DB!