I tried to use Slice and did this into 40 equal area intervals. This worked but the original value (the precipitation value) was lost. However, then I did combined it with the original raster which yielded a raster dataset where all the precipitation values have a number. Then I can use Con("rasterdataset"!=1,"rasterdataset") and Con("rasterdataset"!=40,"rasterdataset") to remove the two 2,5 % tails. Should solve the problem!
Doed this sound reasonable? I dont know if it could be solved it a smarter way, though.