You can probably work around this issue by using hundreds of parallel TCP connections in a round robin fashion.

Assuming you are sending a packet per frame, this setup means that you are only sending a packet every few seconds per TCP connection, which means retransmissions should have already happened by the time you send another packet.