I think we should have a tcpdump of that connection and a feedback and file an SR with support. This sounds interesting. Up to know I only worked on one case where log files took long to be uploaded and the tcpdump revealed a lot of strange TCP errors, which finally were caused by a broken switch port.
I don´t think there is a real chance to diagnose the problem without seeing the network flow, so I recommend to get in touch with Support.
Please let me know if you need help to do so.