Bitcoin doesn't need a fully relational database, it just needs a key/value store . A standalone database like MySQL/Postgres would complicate installation a lot! A library equivalent like SQLite would probably be even slower than Berkeley DB which is what's used at the moment.
Right now the core Bitcoin network has run out of disk seeks, which are a finite resource. That's why some mining pools are dropping fee-paying transactions from SatoshiDice. In other words, Bitcoin has already reached a scalability limit and miners are load-shedding in order to keep the orphan rates down.
LevelDB was designed to optimize for reduced disk seeks as its primary goal, and it sacrifies many advanced database features to get that. Its design is one reason Google BigTable is such a high-performance storage system. Fortunately we don't need those advanced features, but we do need more seeks, which is why it makes sense to use it.
This just reads like joking. Is Mike Hearn a member of core development team? Disk seeks a finite resource? Inform International Union for Conservation of Nature! Probably again the Asian are the culprits and use up the endangered "disk seeks" to produce some traditional potency medicine!
1) Blockchain is stored in a raw file with no structure. On top of it this file is maximally fragmented (for an append only file) due to the way it is written by a mixture of stdio.h/iostream.h calls.
2) The current blkindex.dat is treated like a valueable transactional storage with logging while being just an index, something that a proper database recreates with a "reindex" command.
3) No interprocess locking protocol thus far beyond the single ".lock" file.
I see many years of reinventing the wheel ahead of the Bitcoin core development team. I wonder how many more years until they reinvent the ISAM (Indexed Sequential Access Method, IBM had that since around 1960) and figure out that the index could store unspent transaction scripts until they are spent and purged from the index.
Edit: And how many more years until the discovery here that the transaction key space is uniformly random by design and thus cannot be used alone as key. Then how long until the discovery that transaction space is clustered temporally and widely known generational garbage collection is applicable to Bitcoin?
I see an invisible hand steering the Bitcoin development. It is a hand of an expert in padding the billable hours and lines of code.