The headers message has a size limit, determined by the number of headers, which is triggered at approximately 160,000 bytes (80 bytes × 2,000 headers), about one order of magnitude below the tested message sizes. Logs show the node rejecting these messages, as indicated by the following lines:
Misbehaving: peer=0 (0 -> 20): too-many-headers: headers message size = 2097146 Misbehaving: peer=0 (0 -> 20): too-many-headers: headers message size = 1999996 Misbehaving: peer=0 (0 -> 20): too-many-headers: headers message size = 999996
So contrary to the intent of the test, the tested network messages are no in fact "not_oversized_msg".
In a future diff the misbehavior score increment will be bumped from 20 to 100, and the peer will actually be disconnected, which would cause the removed tests to fail on conn.send_and_ping.
See D16343 for why these tests where added.
Depends on D18025