I first came across this phenomenon about 8 or 9 years ago. Several PLCs connected to the Ethernet network (the Ethernet connection only for remote monitoring/programming - not for control), and all had faulted, shutting down a hydro-electric power station. When investigated, I discovered all the PLCs had erased their application software, looking like they'd just come out of the box.
An IT engineer had been on site at the time replacing a blade in a hub.
On discussion with the manufacturer, this was a known problem, due to what they called an 'Ethernet Storm,' but not a problem they thought was a serious issue and needed publicising. They even had a fix, but wanted £2000 per processor to upgrade the firmware.
I pointed out that there were serious implications especially for Chemical/Nuclear plants etc. and that they should be proactively addressing the problem. They eventually agreed to upgrade the problem to something called a 'code 10' and that way all our processors would be upgraded for free. In the new revision, there was a new register called an 'Ethernet Storm counter'.
Since then, the problem has re-occured and we are now uprevving all our processors to the latest revision of firmware - which they say is now definitely Ethernet storm resistant (we'll see!).
In critical applications where control system components need to communicate with each other, we do use Controlnet, Modbus, etc., monitor for device failures and all the other good practice that Ranjan advocates, but Ethernet is widely used for non critical connections for MIS, Remote Monitoring etc. Who would foresee that a non-critical connection to an Ethernet network could erase the memory of a processor?