Concurrent-reader capacity tests: same file + same plan is the best case

When you measure how many concurrent agents or readers one machine can serve, say whether they all read the same file with the same request plan. If they do, the OS page cache holds one copy of the file pages that everyone shares, and every reader hits the same hot blocks. That measures CPU and private memory per reader under ideal sharing, not a realistic workload. Run at least two more ladders: the same file with distinct per-agent plans, and different files per agent starting cold (copies of one file work, since the page cache doesn't deduplicate across files). Report per-agent memory including file pages, and disk reads. Label each headline with its scenario.