I'm using ucLinux on a Cortex-M3 (NXP LPC1788 @ 72MHz) and it's pretty reasonable. I did it primarily to use Qt as my UI framework, but if I was running in bare userspace I think it would work even better.
The core of the Linux kernel is running out of the onboard flash but the rest is all executing out of that RAM. Kernel drivers, userspace app, and 16-bit WVGA LCD framebuffer. And there's no I or D-cache on this part. =( My app is currently running with memory 75% full.
Just the other day some guy has launched a Kickstarter with a very similar design, but with a bit more breathing room memory-wise:
The trick to Qt on this part was to avoid QWidget and do everything in QGraphicsScene. For some reason it's just a lot easier on the processor. QML is a non-starter on a part this slow in case you were wondering, so Webkit and QScript and all of that is compiled out.