I think the term comes from Computer Architecture/Operating System Design. Working memory in that context is defined as the set of code and data pages that need to be in main memory for the program to function without incurring page faults. As long as the working memory of a program fits within available memory the performance will be constrained for the most part by CPU performance. If working memory set exceeds the size of available memory the program will 'thrash', i.e. excessing disk access that effectively stalls program execution.