We use these adapters exclusively on both server and clients. Everything is connected with the X710s on fiber. Prior to the initial appearance of the issue everything worked properly. When the issue manifested we had abysmal one-way performance AND the multichannel issue, no matter what. We also noticed IO spikes locally (all workstations and the server are with MegaRAID SAS 9560-8i controllers with SSD arrays so we have very high IOPS).
Now with the latest patches we have what I described. We have 4 workstations connected this way, all of them are affected. We just tried resetting the entire network stack on one of them just an hour ago, as part of troubleshooting, with no difference whatsoever. Although our configuration is fairly simple. We disable flow control, configure 8 RSS queues and distribute them properly on cores for each channel, disable EEE and use jumbo 9014 packets..that's about it). In all our tests we get the exact same performance as with multichannel off plus the ramping issue client to server. On all 4 workstations which are identical in every way (both hardware-wise and software-wise).