Was busy with testing RC2, nightly builds recently and can now come back to this question...
I was inspired by the nginx beat which uses the same approach to pull the information. Maybe it's more an architecture decision. IMHO, well defined monitoring interfaces to pull information like http, jmx should be handled by a beat implementation itself. Then you can avoid an additional daemon running on the system which feeds the data into your proposed generic beat with an http input.
Thanks
Oli