Actually, I think it should be pretty much where and how you proposed (except make it a bit distinctive to help prevent adding to the illusion that the node has already been posted).

But I've been burned by my cache losing a node I just barely left too many times and still recall how maddening that is. And I've heard several others complain.

Once the ability to recall your most recent preview gets added, this will reduce the loss in such cases to the changes you made since the last time you push the 'preview' button. But even that isn't something I want to be responsible for increasing the frequency of. Doing so would feel just like I was laying a trap to catch the occasional unlucky sole.

And it sounds like the target="_new" is much more despised than the link is desired. Since my conscience isn't cleared by the vote, I've got two choices left. The consensus regarding those two choices seems pretty clear at this point.

I currently predict someone else will add the link w/o target="_new", maybe even test it and not get bitten. [ This was one of those cases where the browser seemed to sense things and would allow me to quickly jump away from previews for weeks or months without problems, lullling me into confidence that I knew not to go "too far" and then, when a lot of work went into a preview and I was particularly pressed for time, suddenly it'd throw away a preview, but only after I'd gone not quite as far as last time so I'd continue to think I could get away with this if I was just a little more careful. Finally one day it was forced to throw away my hard work (around the 5th or 6th time by then, after years) after a single-hop diversion of only a few seconds right after the preview had been fetched. Then I knew there was no "safe distance" and the caching decisions were too chaotic for me to predict (I could still often load cached preview results from several days ago). ]

But it is still early. I'll wait and see how the signs point...