Coming back fast? Maybe not, I think the surface ice melt (identified as open water) is re-freezing, so that area (already mostly ice) is now being picked up by the data processing of the radar signals.
The freeze may not be any faster than in normal years. I think there was a lot more ice this summer, but it just didn't show up on the instruments.
Some of the speed of the 2008 melt in Sept. may have been just surface melt pools which are now freezing just as fast. (or faster)
Comparing the speed of ice growth in the 1st week of October in the last three years..
2006 59911
2007 64442
2008 101741 = 1 Million Sq kms every 9.8 days at this rate.
That's 60% "faster" than the same periods last 2 years. If I'm wrong, then there is some serious cooling going on. So I think its 60,000 sq kms/day real freeze, and 40,000 sq kms of surface melt freezing over.
Though I note that other years get up to over 100,000 sq kms /day later in October. (either real freeze or surface pools re-freezing)
If you want to check out how the computers "compute" the ice area, check out the this site
(The Goddard data:
http://nsidc.org/data/docs/daac/nasateam/index.html
The AMSR-E use different algorithms for their data, since their satellite was launched about 2002. The Goddard data goes back to 1979, but I think they used two different satellites over that time.
My point in all this is that we have radar instruments reflecting of ice/snow/water/rain and converted by a computer program to ice data based on 1 pixel per 25sqKms. The computer programs make guesses. Every year the ice is different -New, First year, multi year.
So you can't accurately compare year to year, especially in summer. But in any year, the trend from day to day is useful.
Makes for fun arguments tho', whatever "side" you are on in the great debate.