I see, Floating-point numbers can be as large as 3.40282347E+38 and as low as -3.40282347E+38. Integer is less than that, therefore don't have the computational power.
You know if anytime soon they will come out with FP processing plug computers?
Unless what you are looking for is integer processing, then the sheevaplug is not a good solution. SInce its ARM based it lacks a real floating point processor, so anything dealing with FP will be inaccurate and slow in the sheeva.