Even though the second slot is of 16x physical length, only 4 lanes are wired to it while the first slot always has 16x lanes. Using crossfire in an uneven lane configuration, like 16x and 4x, causes unpredictably and often poor performance compared to a system that uses two 8x or two 16x lane slots. You may want to instead go with a single high performing single card like a Radeon 4870X2 or the 5850.
By the way, a single GTS 250 is pretty similar to your current 8800GTS 512MB. The former uses a G92b chip, which is just a die shrunk G92 chip, which is used in the latter, with slightly higher clocks.