Aggregates versus Physical Cube Splits

Hi,

What would be the criteria to decide between aggregates and physical cubes. For E.g.

We have a lot of regional users who are interested in only regional data (say japan, americas, europe etc).

Should we build a global cube and build aggregates on it for the regions or should we build multiple cubes per region and have a single multi-provider on these.

Which would yield better performance ?

Thanks in advance.

Shailesh