If you look at anything in Rasmussen, look at the 7 day average.
OK. I took the numbers from your table.
During the period 9/12-9/18 the avarage was 48.5:45.3 (a 3.2% difference) in Bush's favor.
The numbers in the next 7 days between 9/19 and 9/25 are 48:45.8 which is 2.2% difference.
This means that Kerry REALLY improved by 1% in the last week. These are seven days samples with very small statistical noise.
The methodology of Rasmussen could be bad but it is the same methodology in the compared two periods.