I think it's worth to point out that the work_item.addExpectedResultData() can work with pdg_cachemode parameter to create TOP node automatic read/write cache just like the built-in HDA Processor node. But there's no document I can find contain this information until I read some of the source code of built-in TOP node.
It is a very useful feature in my opinion, so I wonder is the PDG still poor documented yet? Especially on how to create our own TOP node work like the built-in nodes.
Anyway, thanks to the source code, it's really helpful.