I think during the early evolution of nervous systems it was very important to have long axon cells. For example a worm has 3 giant axons which are almost the same length than the worm itself. So there was selection pressure on the length of the axon. The neuron requires a lot of energy to send signals to other neurons. Neurons use energy on nanowatt scale, average human cells use energy on picowatt scale... By long axons it requires less energy to send the message through a single axon than send it through multiple axons. That's the same why we use limited amount of undersea internet cables - currently about 200. If there is more data than a single cable can handle, only then lay we a parallel cable. So a possible answer is evolution of worm nervous systems combined with energy consumption optimization.
Another answer can be that neurons would be never capable to build multiple action potentials parallel. Maybe they cannot consume enough energy to hold charges separated by multiple axons. I don't intend to look for the details of that. It is technically possible to have neurons with multiple axons, so you can grow things like that, but it is not clear if their axons can work parallel or not, and if they don't work parallel does they interfere the communication of each other.
About single axon neural networks there are existing mathematical models. Until there is no mathematical model how a neural network with multiple axons would work, or there is no evidence of it is not possible to build such a thing, you cannot say for sure that it is because of the mathematics of neural networks.
So I have this 3 theories. I think you can find more about this topic by googling a lot, and making new experiments. For me it was enough. :-)