How do you parallelise 1000+ tests to run in under 10 minutes in CI?

Our test suite: 1200 Playwright tests, currently runs serially in ~55 minutes. Target: under 10 minutes for the PR pipeline. I've started with Playwright's built-in sharding (`--shard=1/4` etc.) across 4 GitHub Actions runners. That gets us to ~15 minutes, which is better but not there. Questions: 1. How many shards is too many? We see diminishing returns past 6. 2. Is there a smarter sharding strategy than round-robin? (e.g. based on historical test duration) 3. Any CI services worth paying for that handle test distribution better than GitHub Actions? Would love to see `.github/workflows/` configs from people who've solved this.