A close statistical look at Gallup's
results over time proves clearly that party ID can fully account for the variation in their results.
The USA Today story contains interesting data on party ID and poll outcome. Running the GOP party ID against the Bush-Kerry margin gives a correlation coefficient of 0.73 (the possible range is -1 to +1). This means that the two quantities are very strongly related.
Furthermore, a line fit to the data gives a slope near 1. In other words: On average, every self-reported Republican in the sample added one vote to Bush's margin.
A remaining question is whether party ID varies with what candidate you support. However, since other pollsters weight by party ID and get results that change over time, that is probably false.
I'd say that pretty much settles the question of whether Gallup has biases. The answer is yes!