If it's a truly random sample, which may or may not be the case with Gallup, then the sample's large enough to get a decent picture of reality. That's what random sampling's all about. How many times do you have to flip a coin before you believe the odds of getting heads are 1 in 2?
I have no problem believing 51% think Bush is doing a good job, but that doesn't mean he is actually doing a good job. 77% of the population believe in angels too, but that doesn't mean angels are real. Not that 51% is exactly high praise or anything. The guy just won an election, shouldn't he be popular or something?
I wonder what kind of approval ratings he'd get if someone undertook to actually inform the sample population on a few subjects like foreign and domestic policy.