Differences in preference measurement

There is an anomaly in the public preference measurement and polls regarding the choices in the U.S. Presidential elections between Senators McCain and Obama. Here is the empirical anomaly --

The daily tracking polls/preference measures (i.e. continuous time-series data) appear to show a smaller (1-3 points) lead for Senator Obama. See Gall Up and Rasmussen Report numbers. But the more static preference measures appear to suggest a larger lead (6-9 points, see recent Quinnipiac, CBS/New York Times, ABC/Washington Post, Reuters/Zogby polls) for Senator Obama.

Why is this? Not clear. Some obvious explanations are not plausible. Polls are not taken in the same time periods. No that can be a reason because time-periods overlap. Polls are taken by different organizations. That can not explain it -- for example, the preference measurement by the USAToday/Gall Up (Gall Up organization conducted the poll) found a larger lead for Obama than Gall Up's own tracking poll in June.

So we have to fall on the general explanations i.e. differences in sampling methods, sample sizes, and modeling specifications etc. Sure but these explanations are not quite satisfying because there appears to be convergence in the preference measures both in the time-series data (1-3 points) and the more static data measurement (6-9 points).