Alexa's often wrong on an absolute scale, but why would it be inaccurate for comparing the relative traffic of two companies in the same business?

Because it's completely and totally wrong. I work for Justin.tv, so we have our own traffic graphs. And we also have ustream's traffic graphs, because they use QuantCast. And neither of our charts have anything to do with what Alexa says. They're not even wrong in the same ways.

Alexa is only correct within an order of magnitude, at best. On the other hand, since I have the actual traffic graphs, I can say with confidence we're much bigger than ustream is. So Alexa is accidentally right.