"For results based on the sample of 417 Republicans or Republican leaners, the maximum margin of sampling error is Â±5 percentage points."
Wow! This sample is really very small and hence the results unreliable. In other countries, it is common to ask several thousands of people, and before elections even 10 000 persons. But 417, that's too small. And mathematically, the relative margin of error is larger for small fractions (like Ron Paul) than for big fractions (like Guiliani). That's true even if the sample is perfectly representative.
The +/-5% error means that Fred Thompson might actually have greater support than Guiliani. They need a bigger gap than 10% to statistically say that there is a difference between them.
Are there no REAL gallups undertaken?