In my case I’d like to write software for NXP1769 which is ARM Cortex M3, 64kB Ram, 512kB Flash.
To be honest, to me that sounds like a job for FORTH.
That can be nearly as dynamic as CL (or so I've heard - my own experience is too limited ;), but if you manage to put all (needed! not the whole set) CL primitives into a forth then you'll only need a small translator from CL...
At least it should be possible to put new routines into RAM (via some kind of programming line - JTAG, serial, USB, you name it), and test them from there nearly as fast as via swank (slime)...
Please keep your discussion (and results) on this list, it sounds interesting!