The LIRS papers do a good job of taking different workload patterns, such as loop and zig-zag and mixing multiple patterns together, to validate that their algorithm is robustly good as a general purpose cache. This seems to take traces with similar workload patterns that are distinguished by their excessive size to assert the algorithm's quality. When papers do this it looks good from an academic sense, but it is frustrating from an industry perspective because the workload may not be known upfront or the one must cherry-pick an algorithm for a bespone, consistent pattern. I find it more interesting when algorithm designers try to stress their work and discuss how/why it fails or succeeds. That gives more confidence to those of us who want to implement and use the work in a practical setting.