Apple builds security against those types of attacks into the OS and the SDK. The review process basically adds no additional protection. Holes in the SDK which accidentally expose such features can still be exploited, and those exploits are highly unlikely to be detected by Apple.
It's true that the review process won't catch these problems, but the process increases accountability.
I'm not arguing in favor of or against reviewing though. I'm just pointing out the flaw in Hewitt's logic. There's no magic sandbox that will reliably keep an iPhone app away from the dialer. That's in stark contrast with sandbox systems like Java, where the dollar value of a sandbox break is extremely high.
