Using the current codebase for tomato-rt as of Jan 24th I can replicate this every time. This is also apparent on the K26 RAF firmware.
I'm trying to get a K26 firmware working on this hardware to provide for IPv6 tunneling/networking.
A wget of the centos 5.5 dvd iso was not enough to cause the reboot but a git clone of the tomato tree causes it to happen.
Looking at top it hits around 90% sirq and then the router starts appearing sluggish (slow top refresh). After a couple of minutes the router will reboot itself (full reboot… can see fresh start to uptime).
The git clone got to about 58% of object received (283MB) when the router stopped responding and restarted.