The "cost" is that people are doing things that don't necessarily contribute to the bottom line -- for every GMail, there are 1,000 low-impact ideas.
Isn't that the motivation for the whole 20% idea, though? If you can produce 1 GMail for every 1,000 ideas, it probably doesn't matter if the other 999 didn't produce much of tangible value, because the programme almost certainly paid for itself just on that one success anyway. Meanwhile, you still get to enjoy the morale benefits for all 1,000 staff for the other 80% of their time when they are working on assigned tasks.
I agree. But as an organization grows, the points of view grow. People focus on their niche, and don't necessarily get the big picture. Also, the political environment in a company changes as the company grows, and tends to not reward speculative activity well. If you invent GMail, your boss becomes VP or gets more people/prestise/money/etc, If you invent Orkut, your boss gets to talk about how popular it is in Brazil.
One of the bad side effects of innovative, rapidly developed things is that the support org gets left behind. When the organization is small, this isn't a huge problem. When you're a huge company, the Exec VP of Support has an incentive structure to deliver better support. That VP will lobby for more controlled changes and slower product release cycles.
I've worked in places where most of the organization would be angry if some team invented GMail. They didn't welcome disruption.