Yes, exactly. 3f+1 total nodes where f nodes are faulty means that we need 2f+1 honest nodes to guarantee progress in a classical BFT protocol.

I don't see where in DPOS the message complexity comes in though. It schedules round robin and after 2f+1 nodes added their block, the last block is "finalized" since every other node once agreed on a block that built on top of the block that is being finalized. This creates an implicit quorum. However, this doesn't create any messaging overhead.

Also, I'm not 100% sure if the chain would continue working if 49% of the top 20 witnesses would stop producing valid blocks or if that would cause a chain halt because 2/3 are necessary. I read a few things about synchrony assumptions and under those assumptions you can guarantee progress with f+1 out of 2f+1 (simple majority)