I have not stumbled across this issue yet in support, but it sounds strange. Most likely something is going wrong with the policy based routing. I slightly remember I had a similar issue with a different customer and version 6 in the past, where we had to have a look into the Juniper configuration.
What I found in that case was that the Juniper box was rewriting the destination IP address of those requests. MWG requires that IP address to determine where to connect to in HTTPs. However I am not sure if this is the case in this specific scenario. I would recommend to file a Service Request with support and provide us with a tcpdump, so that we can have a look at what is going on.