There's still the problem for the attacker to get people mining to his (master)node, because until he actually has 51% of the hash, all his blocks will be rejected. And the masternodes ban "misbehaving" masternodes already, if they detect that one masternode is generating invalid blocks they would ban it from the network. The banning is not voted by hashrate, but by the 1000 DRK stake each masternode holds. Seems like an attacker would need 51% of the masternodes as well.
It's not perfect, and it's unclear whether a perfect system is even possible (maybe you already have thought of one, and I hope you have, but it might be far into the future). In the meanwhile, the question was, "would this decrease the mining centralization and the 51% risk?".
You are getting too detailed for what I am interested to explore about DRK. I will just say the game theory is very likely more complex than you've considered. For example take a look at the selfish mining state machine and equations. That was a non-intuitive result that iCEBREAKER would think is impossible until the reality was rammed up his a$$.
In your case, for example the masternodes can collude. Also what ever randomness you have for trying to spread mining shares to all pools can potentially be gamed.
My conceptualization of attempts to enforce decentralization by forcing randomization, is as a Coasian barrier that opportunity cost will find a way around. Random smaller pools are less efficient than larger pools, not only from the standpoint of variance but also from the value of the information and control that can be extracted.
My solution is about offering some value that is greater than the value of larger pools, thus removing the economics (opportunity cost) of the Coasian barrier.