I assume you are using the phrase "child molestors" as a metanym for "Republicans." LOL
Actually, Gallup and Rassmussen are the most constant 'outliers' favoring Republican and Conservative positions among the 15 or 20 top polling organizations. Throughout the Bush administration, both were consistently showing Bush with approval ratings significantly higher than the mean value in a normal distribution of all of the top polls. (Zogby is an example of a consistent outlier in the opposite direction). In Gallup's case, they acknowledge that they include a higher percentage of Republicans in their sample groups than exists in the population at large, which inevitably slants their results. Given that the poll's founder, Dr. George Gallup was a politically conservative evangelical Christian with strong Right Wing leanings, this is, perhaps, understandable.
In the last poll of George W. Bush's performance ratings, Gallup had him at 34% approval and 61% disapproval, tied for the best rating with Fox News (another frequent outlier). Other major polls had his approval rating as low as 22% and disapproval rating as high as 73% (in the case of the NYT/CBS poll) and 24% vs 66% in the Pew poll.
Any individual poll, with or without political agendas, should not be used to define serious trends. A dozen polls, asking essentially the same question in different ways, will come up with differing percentages. Sometimes this is inadvertent, but some polls will intentionally phrase questions in order to skew the answers.
In standardized testing, such as the SAT, if an editor so much as inserts a comma in an existing question, or the typesetter changes where the linebreaks in the question occur when the test booklet is printed, all accumulated statistics on that question from its use in earlier tests are thrown out and it is considered a new item. A change that minor can affect the results.
The imprecision of polling, both in the way questions are phrased and in the nature of the answer options that are offered, means that the only statistically safe way to use polls is to follow a variety of polls on a given topic and regress their results to a mean, perhaps throwing out any outlier on either side that is significantly outside one or two standard deviations.