Actually, my vote goes for #2 .. in terms of sheer number of
prefixes whose routes were affected (obviously a question of
how you define "impact..") Take a look at the following:
These pages contain some unsettling analysis of the effects
of Microsoft worms like Code Red II and Nimda on global BGP
routing instability. They've been significantly extended
since last week, and we *strongly* invite the NANOG community
to send us supporting data (or even anecdotes, let's be
generous) from the propagation periods.
We were shocked to see how little sustained global effect
on routing stability there was from power outages, train
wrecks, backhoe fade, and the like. In terms of generating
sustained routing noise and affecting (at least transiently)
large numbers of prefixes, the worms win hands down.