exponential sublinear curve

I'm not sure what you mean by that phrase. The curve is superlinear at every point (adding 1% rshares/votes adds >1% share of the pool), however it converges to linear.

The entire process is an exercise in balancing various considerations in a dynamic system, not just looking at shifting a few percent of rewards given an assumption of everything else being static.

The curve itself is in fact pretty mild and close to linear overall. Payouts below about 16 STEEM (I guess there are different estimates as this is approximate and I can't say that with some different assumptions 20 is definitely wrong) will get somewhat less than linear and those above will get somewhat more than linear but this is a matter of percentages and is not extreme.

The overall effect is actually reasonably close to the n^2 whale experiment that you liked where there is still a degree of superlinearity but it is muted to the point that the largest stakeholders and vote concentrators can't take effectively (nearly) all of the rewards.

decrease wages 40%

There are no wages being paid here and if there once were (which there weren't, this is a distribution method for a token intended to help the token accrue value) they've already been reduced by 99%. Arguing over protecting the remaining 1% "wages" at the expense of restructuring and rebalancing the entire system to where it at least might work better and avoid another 99% decline is quite pointless.