Lambda the Ultimate

Reading about Constraint Handling Rules, there is a claim that CHR is the first declarative language for which optimum efficiency of implementing all algorithms can be proved, and that this cannot be proved for the pure part of logic (eg Prolog), rewriting (eg Maude), or functional programming (eg Haskell), without adding imperative features.

Should we all be trying to develop languages based on CHR, rather than functional? What other language types to proofs (or disproofs) exist for optimal efficiency?