it all depends on the load of your server.
SubA would fail to SubB (if SubB has enough horsepower to handle the device load)
Typically, you would want to spec out your servers accordingly.
So if SubA had 50 percent devices loaded on it, and SubB had 50 percent devices loaded on it, SubA and SubB can fail to each other.
If Sub B had 80 percent devices loaded to it, and Sub A had 80 percent, and one failed, you would see a huge CPU and memory spike, possibly bringing down the backup server.
In that case, you would want to utilize your publisher and break divide up the devices for proper failover, or simply add another subscriber to the cluster and divide the devices again.