Could you send me a log from a session that suffers from this problem? I don't think hand-tweaking a per-connection MTU is a sustainable solution here. Sadly, this is also mostly beyond the scope of application control. We do have a congestion control system that tries to make sure we don't send too much data if client can't handle it. Our internal MTU starts at around 1300 bytes (Ethernet theoretical guaranteed max is 1500). Forcing it to a significantly lower value can cause issues, however, we rarely send packets that big, we simply do not generate that much data in a frame and can't wait too long for the buffer to fill as it'd affect ping.