I have an initial read only implementation of this setup.

It supports query/display in html, rest and xml interfaces.
You can specify a cache parameter, but using it raises 
NotImplementedError.

It does not support: search, sort or grouping by the computed field.

Doc is also done (bloating customizing.txt even more).

While the initial use case is read only, a developer I know pointed
out that this could be useful abstraction to model external data
sources into roundup. So in theory setting a property in a remote
db (oracle, mongo, ldap) could be done via this mechanism. I am not
sure this is better then using a reactor. It would keep data
retrieval together with data manipulation/setting code but....

I am still using the classname Computed, but I think it needs a better
name. Maybe: Operator, Agent, Broker (my fave so far), Proxy... 
Suggestions?

-- rouilj