Consider to duplicate similar GitHub self-hosted runners in case if repository load will require it.

[shimkiv]

We currently will have 1 self-hosted runner per CPU architecture. This might be not enough because GitHub repository jobs will be queued and executed sequentially one by one. This increases awaiting period increasing overall PR/push checks time required. Moreover if runner won't accept/queue job for some timeout (IIRC it is 30 minutes) this job will be cancelled.
Simple solution is to duplicate similar runners available by the same label and scale further if needed depending on the repository activities/load.
Long-term solution though will be in using more complicated approach with the K8S involved so that we can scale up and scale to zero if needed.