> Have better ways of reproducing this been found?

No. I only see it occasionally when editing C++.

> The most likely part of Semantic dealing with overlays changing is semantic's incremental parser.

An overlay modification hook was a speculation that later investigation revealed was wrong. It is semantic-change-function that breaks the Elisp manual's advisory:

"The functions you use in these [change] hooks should save and restore the match data if they do anything that uses regular expressions; otherwise, they will interfere in bizarre ways with the editing operations that call them. "

> It looks hard to reproduce...

Hard to reproduce, but easy to fix: put a save-match-data in semantic-change-function or perhaps semantic-mru-bookmark-change-hook-fcn .

> It doesn't move point, but it calls out to other hooks.  The tools that uses those hooks I checked, it it doesn't move the cursor either.

It's not a save-excursion that's missing, but a save-match-data.