I don't want to pretend to tell you that I know what is wrong with your embedded system there. However it seems likely to me from what you said... "the chip seems to stall or pause for a short period" is highly indicative of some kind of software bug that you have that sends your code out to lunch from time to time. In the cases where the reset is not happening the bug must still be permitting a code path wherin the watch dog gets re-triggered before your code "seems to recover itself". On the other hand when the thing has reset it appears that the software lunch break has taken too long to supply the watchdog re-trigger and thus you get hit with the watchdog stealing the lunch sandwich (ie the dog bites and you get reset).

Michael Karas