Don't align the buffer_head slab on hardware cacheline boundaries.
It's on the wrong side of the speed/space tradeoff, especially for
P4's.