Regarding point 2.a.i ("adds random latency to packets to reduce bandwidth of timing channels"): There's actually a "better" method. Hash every packet, and add

Regarding point 2.a.i ("adds random latency to packets to reduce bandwidth of timing channels"): There's actually a "better" method. Hash every packet, and add latency not only randomly, but also, for each field in the packet, add latency based on the hash of that field. (In actuality you want a random salt, so, HMAC or similar) That way, assuming they cannot figure out the salt, you actually actively prevent a large chunk of timing attacks. You're adding deterministic variation that hopefully swamps the server-side variation, deterministic being the key word. You cannot get around it by repeating measurements, as the delay stays the same. (Alternatively, keep each packet so that exactly 100ms or whatever has passed between getting the packet in and sending the response, silently discarding any packets that take >100ms to process.)