[Carpet] optimisation of some dist::* MPI wrapper functions

[Richard Patra]

Hi Erik,

> The communication pattern depends on how the 27 processors are  
> arranged in space.  If they are in a plane 3x9x1, then the patterns  
> is different from a cube 3x3x3.  During startup, Cactus prints the  
> global grid shape.  What did it output for your run?

Ok, they are arranged in a plane 3x9x1, I have overlooked this; so I
reran the benchmark in a 3x3x1 configuration. Then during the second
stage of the communication phase a symmetry around process 4 can be
observed. 

Thanks for your hint, Richard.