How do you optimize a large Karate regression suite for faster execution?

Most of the runtime cost in a large suite comes down to how much actually runs in parallel, how much redundant setup happens per scenario, and how much time is spent waiting rather than executing.

• Tune parallelism - increase the Runner's .parallel(n) value to take advantage of virtual threads, since the overhead per additional concurrent scenario is much lower than with traditional thread pools.
• Use callonce for expensive, reusable setup - avoid re-running costly setup (like authentication) per scenario when the result is safe to share across the whole run.
• Replace @parallel=false with targeted @lock tags - only serialize scenarios that genuinely contend for the same resource, instead of pulling them out of parallel execution entirely.
• Use tags to split the suite for CI - run a fast @smoke subset on every commit and reserve the full @regression suite for less frequent runs, rather than always running everything.
• Prefer retry until over fixed sleeps - polling-based waits finish as soon as a condition is met instead of always waiting a fixed, worst-case duration.
• Keep mocks local where possible - testing against fast, local, stateful mocks instead of slower, shared external test environments removes network latency and environment contention from the equation entirely.

The single highest-leverage change for most suites is usually maximizing safe parallelism: replacing broad, all-or-nothing serialization with fine-grained @locks and pushing the Runner's parallel count as high as the underlying test environment (and any real external dependencies) can actually tolerate.