> This is how I understand it: up to step 5 we have determined some
> pivots, which there are P-1 of, and broadcast these out to all
> processors.

Yes.

> Now, we break up the already sorted data on each processor
> into P sections of data and.....this is where I get lost.

Call the pivots v(1), v(2), v(3), ..., v(p-1), which all of the
processors have received.

Each processor then determines its local block of elements B(i), 
for each (0 <= i <= P-1),  such that for each element "b" in B(i),
v(i-1) < b <= v(i).   (Assume that v(0) = - infinity).

> Are the
> pivots supposed to be the min/max values for the data sent to each
> partition? If so, this means that each section of data will not
> necessarily have the same number of entries. 

Yes, that is correct. The number of elements in each block, |B(i)|,
(0 <= i <= P-1), (and on each processor) may differ from block to
block.