-snip-
Hmm, I might be mistaken but I've went ahead to identify the OKEX blocks in questions and here's their stats for both of the instance for which they were mining the empty blocks with the preceding blocks included as well.
For the timeframe in question, the first field represents the received time while the second represents the timestamp within the headers.
664331,-snip-,2021-01-03
18:07:20,2021-01-03 17:29:06,-snip-F2Pool
664332,-snip-,2021-01-03
18:07:38,2021-01-03 17:30:52,-snip-,OKEX
664365,-snip-,2021-01-03
22:41:19,2021-01-03 22:13:48,-snip-,BTC.com
664366,-snip-,2021-01-03
22:41:34,2021-01-03 22:17:12,-snip-,OKEX
The second field is the block timestamp which is by no means accurate.
There are several miners still employing SPV mining to this day even with the optimization made with propagation time as well as validation times. As with most things, without access to see the code behind the mining pool, it's impossible to tell exactly what the miners are doing. The best hunch here is that they're probably SPV mining. Covert ASICBoost would favour empty blocks but given that most miners are doing overt ASICBoost nowadays, it would be highly unlikely.