What about trying?
it on the out-of-sample period 2001-2010
drop the shorts and see what drives the performance
try different start periods or rolling test to see if there is any timing luck involved
see the distribution of alpha each year to see if the models holds or if its just one or two years that drive the performance
I usually use Mod(StockID,5), but I adjust the rank < x since each sub-universe is smaller, the turnover will increase dramatically.
And definitely try AlgoMan approach. With a test of just 25 stocks, the outliers could be a big part of the performance.