I got a bunch of helpful comments, but it turns out

that we have a hardware problem with virtual memory. It

also shows up in the sundaig test of virtual memory under

4.0.3, but it is less obvious than under 4.1. Apparently

the 4.1 and 4.1.1 kernels put greater stress of this

particular aspect of virtual memory.

Leonard Evens len@math.nwu.edu 708-491-5537

Dept. of Mathematics, Northwestern Univ., Evanston, IL 60208