Just an update. I am quite taken by Charley Shattuck's myforth (click link and scroll to the bottom of the page). Porting it to an obscure 8051 part (for 24 bit sigma-delta ADC) greatly increased my knowledge of 8051 assembly. Myforth is more of a Forth-based macro-assembler than a Forth dialect. It is both minimalistic and rich.
I got great pleasure out of implementing a moving average over 24 bit ADC samples using just 8-bit operators for the math. It reminds me that I take a lot for granted when I toss about 32 bit integers using C on an MCU.
(I've previously used SwiftX Forthfor 8051 programming, but the part in question had only 768 bytes of RAM -- not quite enough.)