My uneducated and totally intuitive suggestion for people having problems during swapoff is to enable swap on microsd first, i.e. "swap refreshing" procedure should look as follow:

swapon mmcblk1p2
swapoff mmcblk0p3
swapon mmcblk0p3
swapoff mmcblk1p2

Of course You NEED to have a swap partition on Your SD card, but it may be even just few megabytes. Going back to my totally unchecked and not proofed (cause I'm unable to replicate that sort of reset when swapoff) ideas, i think that swapoff hard way without enabling swap elsewhere may drive some critical processes mad - but only in certain circumstances, i.e. not always reproduceable. then, of course, watchdogs perform reset.

By the way, and what may be even more important - using swap (normal 768 Mb, not few-megabytes variant only for refreshing) exclusive on microSD and tweaks as discussed on thread mentioned by qole (last pages, first post is outdated), I got 0 problems with reboots/crashes/whatever, no matter how big file i test.

anyway, i think that it may be worth to mention (again in topic about I/O improvement) recent discoveries about MyDocs exclusive involvement in I/O problems - maybe that is something with vfat module or whatever related to vfat? I wonder if someone who got MyDocs reformatted to ext3 - for example - got the same problems.