Okay, thanks.
So, I personally would recommend the increasing of Alphas. Even in the same machines. Maybe 9 Alphas (so you have 3 groups) - so each machine would have 3 Alphas. After that, you could setup your client (Java client?) to distribute each transaction between these Alphas. Or you can use an LB (NGINX) here an example Dgraph + Nginx Load Balancing · GitHub
That way I can guarantee that your mutations would be fine over time.
If you don’t wanna increase the number of Alphas, you can just try balancing at the client level.