I've seen a case where a single error in the configuration file of a $VENDOR_1 router was accepted (due to an 'undocumented feature'), and this caused the wholesale importation of BGP routes into the IGP, which caused most of their $VENDOR_2 hardware to spaz out. Locating the single error was a matter of hours, not minutes, so effectively a typo took out that ISP - and it's considered by most to be a relatively well-designed network.
I have also seen a variation of this where the boxes which got flooded by large IGP tables run out of memory and not recovering (because there was no memory left) after the broken router was withdrawn. Eventually the network got fixed by restarting every box in succession. Not sure if there are safeguards for this now or if everybody buys all their IGP routers with 512M or more. Pete