Forcing WSARecv to overlap

I'm triying to implement a server which uses IOCompletionPort to read from its clients. I have something very similar to this example.

If I understand correctly, this should be my design:

1. [Main Thread] Create a listen socket, bind and listen
2. [Main Thread] Create an event and attach it to the socket accept signal using WSAEventSelect
3. [Accept Thread] Wait for the event and accept a client
4. [Accept Thread] When a client connects, use CreateIOCompletionPort to use the IOCompletion queue with it
5. [Accept Thread] The accept thread calls the first WSARecv with overlap parameters
6. [Worker Threads] Use the queue to implement a leader-follower pattern on WSARecv

After some reading on WSARecv (Here) I discovered that WSARecv might return immediately with the data if the data is ready. That seems kinda weird because that means that a worker could loop on WSARecv without returning to the IO queue if the client sends fast enough - And that could cause client starvation...

Hereby my questions are:

1. Is there a way to "force" WSARecv to not returning immediately? I mean, to return IO_PENDING 100% of the time?
2. If not - What would be the correct design, optimized for scalability?

This is how I use WSARecv:

flags = 0;
receiveResult = WSARecv(clientSocket, &(olStruct->Buffer), 1, &bytesReceived, &flags, (OVERLAPPED*)olStruct, NULL);

olStruct is an extension for the OVERLAPPED struct.

EDIT: I ended up reposting what I got from WSARecv using PostQueuedCompletionStatus. I'd love to hear other solutions though See Answer

It seems that when WSARecv completes immediately it returns the buffer AND posts to the IOCompletionPort queue.

So the current design still stands.