Lambda the Ultimate

There's an equivalence between partial application and staged compilation. Are there any languages that took this to an extreme, and used runtime compilation pervasively instead of closures?

More than likely, this naive approach will actually be slower than closures due to I-cache flushes, but partial applications can be viewed as staging hints, so I'm looking for any work along those lines. A sort of unified theory of partial application and staging, ie. when it's appropriate to generate specialized code instead of a closure, or something along those lines.

Deferred Compilation: The Automation of Run-Time Code Generation is along the right lines. Anything else?