However, with new k/v stores and other database models, I don't see why we need to mangle SQL to work with every single type of data model. If it can be mapped, great. Otherwise, let each database build a query language to truly exploit its power.
There countless paradigms that prove my point. The web won out over the desktop. Relational won out over flat files. XML over other formats.
If the paradigm of NoSQL will succeed, customers will want some belief that they can ultimately escape the clutches of a particular vendor's product.
The danger here is a proliferation of syntactically similar but semantically different query languages, and mass confusion as developers attempt to switch between data storage providers.