is everything running on a single VM and you try to this way optimize the usage of this single VM? perhaps switching to something like a serverless container such as with ECSTask or CloudRun infra blocks could solve the root problem without adding this work queue complexity?