One of my stress runs on the mirror branch hit a mismatch bug. Using that specific CONFIG I've been able to reproduce it twice in under 100 iterations.
You must use the wt-8372-mirror-format-objects branch (unless it has been merged by the time someone looks at this) to use the CONFIG.
The base table is a row-store and the 2nd mirrored table is FLCS. The first 4 (of 9) tables are mirrors.