In a lot of scenarios, the latency is not much of an issue, but on wlan/3g, this very fast become very noticeable. In the few cases since, I've used ssh on wireless, I suddenly find myself noticing how much of an issue the sudden feedback delay might be at some times.
The only obvious downside to Mosh I've seen so far, is that a lot of enterprisy firewalls tend to block UDP traffic.
Perhaps this extra complexity would lead to all sorts of strange bugs though.
I don't think ssh sends that back to the client terminal though, but I must say I'm not really sure how that's handled. On one hand I'd like to know better how unix TTYs work, but on the other I value my sanity so I'm probably not looking back into those just now :).
In that case, I guess this functionality would be implemented similarly to the example you mentioned of stopping the client from sending every keystroke at a password prompt, right? It seems like it would be strange for ssh to send the 'toggle echo' signals to the client, but not the 'toggle canonical mode' signals (which the client could also handle in a useful way).
TIOCPKT_IOCTL(EXTPROC) and TIOCSIG support landed in Linux kernel 2.6.36: https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux....
If you change to "local echo" it would cause problems with programs such as vi which will interpret certain characters (such as hjkl) as special commands and won't actually draw them on the screen.
If you start a remote program that just "eats" your input without updating the terminal (like a password prompt) you won't see the 2nd tcp roundtrip.
You'd have the same behaviour with a serial link, rsh, telnet or anything else.
I think I saw a custom SSH client on hn some months ago that would try to compensate high network latencies by echoing the characters locally while waiting for the remote reply. It would then undo the local echo and use the "authoritative" reply instead.