1. I don't see how it adds complexity, it's just a filter that you run over a random number to pick the next one, the current one is more complex because it has the multiple tiers including the miner (which is going anyway) and the backups and they are already distributed using a curve like this, so I don't see how this is adding but rather would be eliminating.

2. Political behaviour is unavoidable anyway, and nothing in how it is at the moment stops people buying votes. Really, you raise the subject of vote weight calculations there, because there is a limit of 30 and how does that work the more votes you fill? Are they equal weight or does it distribute your vote weight like it does in the forum? I think I should know this, I know, but though I know SP weight is used, it is not at all clear to me how it distributes (not voting is a wasted vote?). 

    In my opinion votes should consume and provide to the total for the candidates an amount divided by the number of candidates voted for each, or the proportioning be assigned as part of this vote to bias your vote distribution. Perhaps this is an area that needs more attention in fact right now.

3. Dropped blocks: When I see multi-thousands of dropped blocks and consider it's at a bit over 9 million blocks now, this says that probably somewhere around 0.01% of blocks are probably dropped on average, or somewhere around 8 a day. You are probably right, it isn't a big enough problem certainly not right now. 
    
    But will this problem become bigger with a larger number of witnesses operating? I think yes, so perhaps it's more of a 'something to consider in the future', but also it fits with the scheduling curve I suggested which will scale out with increased network capacity, but due to the shape of the curve the spread of the block slots amongst candidates still stays sharply in the bottom 2/5ths (er, top 2/5ths) of the ladder. Most of the 'excess' witnesses would be way down the other end and still not get so much that the change in block rates for higher voted users would be that great.

I think the thing of assigning two blind pairs of secondary primary is perhaps adding unnecessary complexity but it probably will help with bigger network load where there likely will be an increase in latency and consequent increase in block misses (is 0.1% or 1% or 10% the drop rate that we describe as a problem?). Like I said, it also would get backups to keep their stuff running well or they miss their stuff and it's taken by those who do.

My ideas relate to the future where the chain has to be built in parallel to maintain the low latency of transaction clearance. They wouldn't be as beneficial right away I think. The witness election algorithm I think may need to be looked at more if your concern is gaming the system, and the forum is a ripe proving ground for self-voting behaviour and vote gaming systems to see how they have been handled and where they are relevant to witness election. 

I don't remember reading much anywhere about witness election gaming yet but since you raise the subject I am now very curious.