Not for OLAP, but this seems to be about OLTP. If you're trying to sort 10GB of data for a traditional OLTP query, I would consider that "too large" and need of some data structure optimization.

Also it's specifically about queries where the sorted data is too big for memory:

> The final change is aimed at larger scale sorts that exceed the size of work_mem significantly. The algorithm which merges individual tapes was changed to use a k-way merge. When the number of tapes is large, less I/O is required than the original polyphase merge algorithm.