What I said stands.  'BP CG' for MPI divides the images into sets to
  that are sent to each node. The images from a set are loaded into memory
  then that memory is sent to the node.  If there are too many images the
  loading (allocation) will fail.  Sometimes it fails on the allocation and
  gives a nice error message, but sometimes the stack will overflow in a
  function call and the error will be not obvious.

  It would take some experimentation to find out what number of images of
  what size could be sent to a single node without failure.  I say
  "underterminable" because this will be difficult to due algorithmically
  since it is dependent on: number of nodes, image size,
  differences in node memory, and previous stack usage.

  Currently one of our projects contains  about 125K particles in TOTAL.
>