Ask HN: Is HTTP 2 problematic for mobile or NATted clients?

I suspect that HTTP 2 (and SPDY before it) has a subtle but severe performance issue that primarily affects clients that roam between IPs or that are behind NATs that are restarted. Specifically, I think that existing HTTP 2 implementations are bad at destroying nonworking connections, leading to seconds to minutes of lag.

The symptom is that sites like gmail sometimes become unresponsive for extended periods. I think this happens when the client sends a request over a TCP connection that is no longer functional due to router reboot, NAT timeout, roaming, or anything else.

Telling Chromium to reset all sockets using net-internals immediately makes everything responsive again. Firefox has no such mechanism that I can find. Neither browser seems to have any good diagnostic screen to show HTTP 2 connections, the current requests using then, and when data was last received on the connections.

Am I right? Is HTTP 2 to blame? If so, how could this be mitigated? (Something using ping frames, perhaps.)

My best guess is that, with HTTP 1.1, keep-alives and pipelining never worked particularly well.