Running off 90% of the authors (as well as non-author users) is mostly a function of the value declining by 99% and the overall value proposition being weak IMO.

That being said, if hypothetically 90% of the authors did leave, the rest would see their rewards increase by 10x. It is self-correcting in that sense.

That superlinearity is now down for steem poor people instead of up for steem rich people

There is no difference. That is the exact same thing stated two different ways.

That only the authors that get less than 20 steem experience the superlinearity

This is not true. Every single vote has non-linearity (whether we call it super- or sub-linearity). The cutoff point of 16-20 is a rough estimate of where things will be similar in some sense to the status quo, but this depends a lot on how voting behavior may change,

As I noted earlier, the degree of non-linearity is somewhat modest. Even at the very bottom (near zero), the relative value is something like 50%, which is much much less of a non-linearity than existed under n^2, even with the whale experiment. At modestly less than the 16-20 range, say 10, the non-linearity is still there but it is much smaller. I don't know the number but I might guess at say 10%.

How much sp is going to be required to make a solo vote larger than dust on a comment?
That number is going up alot, yes?

As I understand it, something in the neighborhood of 2x. Solo votes aren't really the point though. Voting is supposed to be a consensus process. i.e. Go ahead and put your vote on there even if it doesn't reach non-dust. The effect of that is another minnow can add their vote and perhaps get to non-dust, which they wouldn't be able to do alone. If the comment is truly meritorious, we should expect more than one voter to think so.

Solo voting is almost synonymous with milking. It's what we have to get away from, not what we have to protect.

Overall I think we will have to see how things work out including the knock-on effects of more downvotes (which redirect rewards from that being downvoted to everything else, including potentially low-vote content, offsetting some of the above effects), etc.