Thanks, this was the only reply I got that gave me a decent explanation. I guess it all boils down to defining business size by the amount of profits rather than by revenue or assets. The article's claim, then, is contingent on the cost of a single unit of innovation being set by this size metric and not the others. I'd say this is a very non-obvious assumption, but at least I now understand the claim.
Can anyone give an argument for why the cost to reduce costs by 1% should be a function of profits rather than of revenue or assets?