Here is complety alternative approach to your prob...

# coroutines

[elizarov]

Here is complety alternative approach to your problem. This pattern is called "worker pool":
1. Create a channel for your requests: val requests = Channel<Request>()
2. Launch n coroutines to process this channel:

repeat(n) { 
    launch(context) {
        for (request in requests) 
            request.execute()
    }
}

Done. You have a guarantee now that at most n requests are processed concurrently.