Currently the network has the risk of validators taking control by gaining too much voting power.
This can come about by having really low fees, or by exchanges using IOUs for trading while staking the users coins and using these as voting power.
By splitting the voting power from the rewards, it would ensure that validators will still get the same amount of rewards as they normally do, depending on how many atoms are staked with them. But the voting power will have a hard max limit. I propose (num validators / 10) so that no one validator has more than 10% of the power.
This does not stop a validator splitting into two or more to try and get more voting power, but this would signal to the delegators that the validator is looking for power and not just the rewards, which would indicate potentially neferious purposes.