All of these proposals that change policies and requirements where the actual effect can't reliably be projected (i.e. what specifically will change with increasing the requirements for autoconfirmed, how many people who vandalize after being autoconfirmed now will simply wait a little longer or do a little more) ought to have a built in expiration clause. Run it for a month, or two months, and if the community doesn't ratify it as being useful then it stops on its own. Otherwise these things can be implemented, run without achieving a tangible benefit, and never be stopped because doing so requires a huge effort.
Rollback rights come to mind. Whether granting it at all and doing it the way it is done has worked out at all or better than the alternatives is something no one has studied - the people who disagreed with it (many people) mostly ignore it, and its only newbies and admins in favor of the procedure who ever think about it. How about adding Twinkle as a gadget? For editors who "abused" twinkle, you could remove it and project their .js page. When it was added as a gadget, that ability (I assume) went away. What are the effects of that? If no one organizes a comprehensive review no one will ever know - and even if someone does, and there are significant downsides, changing it back requires an act of Congress. So changes like these should all be forced to undergo a probation period to give concrete verification of their supposedly positive effects.
Nathan