weird CPU untilization of different operation to the same cube

[ricky_ru]

The hardware is pretty powerful. 4 CPUs with 16 GB memory. There are 40 cubes each of which is about 100MB, 3GB totally.  These cubes are partitioned as time-based.

There are 4 dimensions. d1,d2,d3 and time.

I draged the time dimension first so that the query will be pointing to only 1 cube. Later I drag the d1 and d2 respectively. The difference of d1 and d2 is d2 has more values(36) than d1(6) at the first level. But you can check the CUP untilization is not as I expected that the caculation for d2 should use more CUPs than d1.  But the fact is that d2 operation only uses 1 cpu while d1 uses 4 cpus. And the result is d1 only takes 8 secs to finish while d2 consumes 30 secs to be done. Please check the pictures.

d2 should untilize more cpu resource to do the caculation. Does anyone has clue on this? Is there any documentation I can refer to for powercube creation and performance tips?