I have been looking into writeback so I have edited the Adventure Works cube
to make it writeable.

If I execute the following statement the writeback succeeds within a few
seconds.

UPDATE CUBE [Adventure Works] SET ([Measures].[Internet Order Quantity],
[Customer].[Customer Geography].[City].&[Brisbane]&[QLD])= 400
USE_WEIGHTED_ALLOCATION

If I then execute the next statement then the query will run for about 10
mins before timing out and the memory usage will skyrocket.

UPDATE CUBE [Adventure Works] SET ([Measures].[Internet Order Quantity],
[Customer].[Customer Geography].[City].&[Caloundra]&[QLD])= 400
USE_WEIGHTED_ALLOCATION

If I let it time out I get this error "Memory error: Allocation failure :
Not enough storage is available to process this command.".

However if I commit the writeback changes after the first statement then the
second will execute within a few seconds.

Is the problem something to do with the fact that the members of all the
other dimensions are using the default all member and so the writeback must
assign values for many cells? If so is there any reason why the first update
would succeed so quickly?
Should writeback on non-leaf levels be confined to very small cubes?

Thanks
Chris