You can absolutely kill an Arduino or the ATMega328 or a pin or two. Many people have done this. I have heard many more stories of people killing pins than killing a whole 328, though. I've overheated my Arduino's regulator a few times and it has shut itself down which is better than I can say about TLC5940 chips.
If you damage the chip, a replacement is about $5 for the chip itself, with the bootloader included, so it really isn't the end of the world. All hail the DIP package, pop it out, pop a new one in. Couldn't be any easier.
If you never kill a chip you haven't been trying hard enough. Seriously... Everyone kills a few chips.
Yeah I've killed a few myself. My only point was that it isn't quite as easy as just an occasional intermittent short. It's the real abuse, shorts to higher voltage or other "sins" that really kill the parts. I agree about replacing a $5.00 chip and your statement about anyone who hasn't... but from my experience an accident once or twice isn't as bad as it seems I have 6 Arduino's; Uno's, Mega's and Pro mini's and I have yet to lose one, perhaps more from luck than anything else and I DON'T recommend that shorts be not considered bad for the processor, Not at all. Just that IMO and experience that most aren't immediately fatal unless you involve the protection diodes, that is always fatal unless current limited and still not a great idea. I participate to the best of my limited ability here every day and although there are many "blown" chips the cause isn't generally connecting an LED W/O a limiting resistor or a relay that requires a hundred mA to operate. This "feature" was my factoid about the designers having chosen well. I've popped a lot of old Mos devices due to shorts and made a whole gang of similar mistakes in the 40+ years I've worked in this field... some because of ignorance and some from poorly designed parts and I guess I've been extremely lucky with mine. I have yet however to loose a chip on an Arduino board. As an afterthought it wouldn't have hurt much to put a 470R resistor in series with each I/O pin... wouldn't help the protection diode issue but it would sure protect the ports from overloads or shorts and IMO if 470R is an issue then the design needs to be re-evaluated... IMO Strictly