Greg Galloway is Testing cube performance improvements

 

Conclusions

• It is clear that ProcessAdd is worth the extra effort if you need to further optimize ProcessUpdate dimension processing.
• It is clear that AttributeHierarchyOrdered=false on high cardinality attributes makes a significant difference to dimension processing. Test case H shows that AttributeHierarchyOrdered=false on high cardinality attributes can double performance of ProcessAdd.
• ProcessUpdate is by far the most expensive option available. It is so much more expensive than ProcessFull that you should definitely verify whether processing scheme #1 actually performs slower than processing scheme #2 on your dataset.

Source: Greg Galloway