we’ve been working on a 2 pager definitive performance best practices doc based on experimenting with various scenarios.  One observation seen has to do with # of measures on the column shelf.  When exceeding 6 measures (some are materialized in the extract, some are window calcs, some are LOD), the processing seems to exponentially slow down and upon studying the performance recording it seems to change behavior to compute each individual cell instead of processing somehow collectively.  Has anybody else studied this and is there any truth to this, is there any hidden meaning?