Testing phones...
Not quite true... yes, you use an anechoic chamber because you are trying to reproduce the effect of "free space" in a constrained environment - ie avoiding the impact of echoes (reflections for RF), diffusion and all those other wave phenomenon that happen in a room that is physically not "infinite" compared to the wavelength you're looking at.
But... the phone does NOT "sit in a non-conductive cradle in the middle of the room". That's ONE test scenario. My employer, along with every other handset maker and mobile operator in the world that wants to test phones properly, ALSO tests with simulated hands (water bags, more or less), simulated heads (ditto with a mannekin), windows (glass sheets) and walls... at different angles, signal strengths etc etc.
That's why you build a $100m. suite with 17 chambers - there's really quite a lot of testing to do...
FWIW, my guess is that the testing DOESN'T simulate the different levels of conductivity of the skin (ie some people "short" the external antennas more than others), which you wouldn't see at all in a phone with an antenna buried inside the case. Even if generally, the iPhone 4's external antenna gets a better signal than an internal one, salty sweaty people shorting the phone across the gap will have a bigger effect than they would just as acting as "bags of water" on an internal antenna.
Equally, the SOFTWARE controls how the phone responds to signal changes. Anyone who says that RF physics means that no software update could "fix" the issue is just wrong. A software update can't increase the signal strength, but it can certainly change what you do with it. You could detect the "shorted" antenna and re-work the baseband signal processing on the fly based on new tuning parameters.... if you had to... It's not as if the only antenna that works is one that is some precise number of wavelengths long, after all...