For one, that's a crime against statistics and formal logic itself. "90% of RAM is not fault tolerant, of course men are going to assume it's not RAM if it's fault tolerant" would be a ridiculous thing to say, because you just can't reason that way.
The kind of reasoning you want to do works like this: 90% of developers are male. Given that there are about as many men as there are woman, a randomly chosen woman has a 10% chance of being a developer. That's valid statistics. The problem is that this works relatively good while speaking to people in a subway, but at a tech conference you don't have a fair sample of either males or females. You aren't choosing at random but from an extremely skewed sample.
Of course there's instances where these kind of assumptions are valid, but most often (and with most stereotypes) it's just people applying statistics wrong.