Pipe and filter could be more scalable because PAF supports the development of a reusable code element Library. The library has no upper limit to the number of parts, and the baseline parts are independent of each other. So new features can be added to the application with more ease than other architectures. Especially architectures where parts depend on some knowledge of the parts above or below them in a layered hierarchy.

However, as a PAF application develops a library of devices made from base parts, the parts lose their ability to change. So, a new feature could be added, but it could be enormously difficult to make a change to previously existing parts.

Also, PAF has performance issues (duplication of many parts)that inhibit it's scalability.

It also is difficult to make interactive because it doesn't have persistent storage.

maybe I was not clear in my question. By scalabilty I mean performance - that is adding more hardware we get more performance. In that regard pipe-and-filter architectures are more easily can be converted into distrbuted ones and thier processing can be performed in parallel.

Pipe-and-filter architector can be scaled by creating several instances of a processing filter and distributing data among them. (Parallel processing)