Choosing Between RSpec's Lazy let and Eager let! to Minimize Latency in Concurrent Test Runs

The goal is to select either RSpec's lazy let or eager let! helper so that test suites exhibit minimal latency when executed under simulated concurrent request loads, while preserving test isolation.

Using let defers setup until the helper is first referenced, which can reduce per‑example CPU usage in parallel workers but may hide ordering dependencies that surface only when workers share group instances. Conversely, let! forces evaluation before each example, adding a fixed overhead per test but making state visible early, potentially increasing total runtime under parallelism.

Which helper yields lower overall latency when specs are run with parallel_tests across four workers? Does the eager evaluation of let! introduce measurable flakiness due to early state initialization under concurrency? How does the memoization scope of let affect isolation when example group instances are reused by parallel workers?