Hi there,
Well, I'll be brutally honest -- I don't think I would go that direction. We didn't pick a database project last year because a majority of the mentors agreed that the 'dsl-based' approach was more of a toy than reality. The main reason for this is that in 'real-world' applications queries are often dynamic and must be built at run-time. So, before we get ahead of ourselves and think about 'sql embedded in C++' we need the basics of getting database access.
At the point the leading candidate for boostification of basic db access is SOCI:
As I understand the state of affairs, the main barrier to Boost submission is getting the code into the usual boost structure, writing additional tests and such. I"ve cc'ed a couple of principles on SOCI, so they may want to comment on my perception. However, if I'm correct, a great SOC submission may in fact be to Boostify SOCI. It's not as 'sexy' as the embedded SQL, but I, for one would be a big supporter.
I like and use SOCI, as well. Although, today I found out about litesql and I like their approach of generating c++ code from xml based table definitions. So, I think a cross of both would be nice. I did my own code generator on top of SOCI but not with all the bang that litesql have. The good thing about generators is that you would deal with sql statements a lot less, since they can be generated by the code generator. The C++ on SOCI generator might be a nice project for SoC. This reminds me: I always wanted to ask if there is something similar to Javaspaces ( http://en.wikipedia.org/wiki/JavaSpace ) out there for C++.