Well, for our current project which is almost done, we had to complete several sequences in dome master format (3200x3200) so we used FXTree to composite our passes.
It was really efficient to do our (not too complicated really) compositing with FXTree
but since our available machines were not of equal horsepower (old single core and new dual core xeons) we had some machines finish first and stay idle, while some others were struggling with their FXTree share.
For example, my "own" workstation (single core xeon) needs about an hour for 250 frames of a given sequence, while one of the dual core ones finishes it in 20 minutes or so.
So, in order to load some more frames to the new pcs we had to stop the old ones and rearrange the frames sequence jobs, something a bit frustrating and riskfull (to lose some frames during the FXTree stop and rearrange).