At first glance, module support would appear to make sense, but I have been thinking about it some more, and I suspect that it isn't worth it:

1) We'd have to define where the modules were kept on the filesystem. IIRC, they are usually in /lib/modules/version. This would have to live on a disk partition as it's size would be indeterminate, so a custom mount would be required. This isn't a big issue, but it'd be a pain to set up.

2) We'd also need the necessary modutils. insmod, which is symlinked by lsmod,rmmod and modprobe is 94k. This could probably live ok on the disc partition too, as long as it wasn't needed before pre-init. We'd also need /etc/modules.conf and /lib/modules/version/modules.dep (about 56k on my system, but it is build specific I believe)

3) What else do we really need in the kernel? I can't see any need for any other filesystem support - ext2 and FAT should cover it. Even if people are using the IDE bus in unexpected ways (eg this compactflash idea - way cool), chances are that anything they plug into it is going to be either ext2 or FAT. Maybe it'd be nice to have ext3 with it's journalling capabilites, but AFAIK this is only available in the 2.4.x linux trees, and hasn't been back-ported.
As for hardware, there isn't anywhere to plug any traditional hardware into. No PCI, no parallel, no USB master, no VESA, no AGP etc. All the current hardware (serial, LAN, USB slave, display board etc) is already controlled by the kernel anyway, so there's nothing to add here.
What does that leave that we might want?