One additional step that sometimes helps: fix the data. You will have to check whether this applies here. Depending on how the original data was created, it's possible that character variables use much more space than they should. For example, perhaps a variable named GENDER should be "F" or "M" but is actually defined as $200 characters long. That's a problem. Moving around data that is 200 times bigger than it needs to be takes significantly longer. Compressing the data can reclaim the storage space, but actually takes longer to process. Each time you use the data it needs to be uncompressed, adding to the processing time. Also worth noting: even if the original data set is compressed, that doesn't mean that the subsets will be compressed.
... View more