dynamic processor grid

Hello,

I have a simulation of a process of sedimentation. Initially the particles are distributed over the whole simulation box but over time they concentrate in one end. To speed up the simulation I have two ideas/questions concerning the effective usage of processor power:

1) Is it possible to use a kind of "dynamic" porcessor grid i.e. to stop the simulation, change the grid and restart the simulation without an external script that uses several input files?
2) does the grid have to be regular or can one processor be applied to the whole empty volume while the other processors build a finer grid over the interesting volume?

Thanks for your answers!