I guess it is all depends on how your durable functions is implemented.
The Azure function app runs on Azure web app sandbox which can scale out up to 200 instances as far as I remember and it has a maximum 300 bound sockets outstanding at a time, which I think it is a sum of both in-bound and out-bound connections. And once this limitation get exceeded the Azure function scale controller will refuse to add more VMs even number of current VMs is way below 200. I think this applies to orchestrator function as well.
So I guess there isn’t a simply answer to your question except things to be aware when you implement your function, you want it to be executed and complete as fast as possible, therefore function instance can be free to take in more requests and you want to be careful and calculate how many out bound requests a single trigger fans out, so it doesn’t exceed the cap.
Regarding Azure redis cache, isn’t something I have experience so I can’t comment anything on that.
Hope it helps.