Let me agree. For our target flexibility and reusability is more important than performance.

Hovewer, AFAIK Cache' has an interesting technology called ECP allowing to separate data server and applications server while using the same database and code. Code is downloaded from database and actually runs on application server while data server only serves data I/O and transactions.
Yes, this does not optimize whole system performace but at least allows to separate middle-tier from database-tier without any change to application code.


>Well, then we have to define what the database is.
>
>If we define the database as the tier, which incorporates both data support logic (data access, transactions and integrity), and the application logic - objects, methods, properties, inheritance - then your approach is right.
>
>If we define the database as a highly optimized data tier, which supports only data access, transactions and integrity with no compromise caused by other functionality, then I would say that your approach is not perfect - lets middle tier to do the objects, methods, properties, inheritance.
>
>So, after all, it is about what meaning we will put under the "database" defintion.
>
>However, working on projects which require scalability as a feature, I found that the approach to have highly optimized/specialized data tier, and separate speclialized middle-tier - are the only way to build the systems with current state of technology which could handle significant load.
>
>For smaller projects, I agree that combining middle-tier with data tier might also bring some cool results. However, at expence of scalability.