There's no programmed in limit, but in practice it very much depends on the HW and a lot on the network that hosts the latency-sensitive cluster communication. Network communication grows overproportional to added nodes.
That said, we know of setups that run a cluster with 51 nodes, but IIRC with very good HW, network engineers and some fine-tuning. If you don't spend too much attention to tuning and "golden HW", but use a separate physical network infrastructure for the cluster network, then going up to 20-25 nodes should be possible without much headache.
Modern server with the current high-thread count CPUs can provide a big amount of compute and/or storage resources; so a handful of such servers can handle a huge amount of load.
What's your use case and how many nodes of which HW do you plan to have?