Most published sets of data are uniform in structure, coming from tabular,
object, or document stores. It's quite optimized to send the data in simple
JSON structures along with a map or profile which adds web scale uniformity
and semantics to it, one map for many objects, a map which can define that
some property correlates to a specific uri along with its constraints,
range, domain, type etc.
In this way you can have interop and ease, optional complexity and
conformity, handling all cases. You just need to separate the two parts,
much of what we programmers and web devs do is just data transformation
anyway, indeed much if the web is just data presented in one form or
another elsewhere, so I'd argue that having plain old JSON with some tiny
spec which allows the complexity to be punted in to shared documents would
be beneficial to all, and lighter to process, and lighter over the wire.
Been saying this for 15? years though.