I understand the difference between 'hot-swapping' and 'liveness'. However (admittedly, I may be mistaken) I believe the Smalltalk architecture already has the requisite functionality (eval and reflection - just like Lisp) to support this, although no one's actually implemented it yet. (And it ought to be more straightforward than building it in a Lisp-to-Javascript compiler on top of a Lisp on top of the JVM!)
Any Turing-complete language has the requisite functionality to support liveness, but something like Time Warp support by the OS/VM would make it easier. But honestly, at this point, even designing the experience (vs. implementing it) is hard enough, and we owe a lot to Bret Victor's talk here. Hancock's thesis sets high standards on how the feedback must be comprehensible (as opposed to some random lights flashing on and off).
I wrote a lengthy post in the history article you linked to. Its just not the live programming history that I'm familiar with, they seem to be falling into the same smalltalk mechanism trap that I was talking about in this thread.
Some of the basis for my assertion came from this article: http://liveprogramming.github.com/liveblog/2013/01/13/a-hist...