Is this more about new language constructs or a better run-time execution/solver ?
New language constructs. The 'WAM' in the name refers to the Warren Abstract Machine, one of the more efficient ways to implement Prolog in a procedural language.
Also, using the SICStus interface and semantics for attributed variables should make for some very general constraint solvers, yes.
What do you think about answer-set programming vs more general solvers like SICStus.
http://github.com/mthom/rusty-wam