Thanks for your question.
I do a quick rerun on the latest code (v2.2) and similar result is received. I also took a look on the transaction. Seems that the readset only points for a block# and tx#.
I don't think committer (any peer receiving the new block) can tell whether the modification is an intended one (done by endorsers) or tampered one. From the writeset it only shows the final result, which already contained the tampered data.