Couldn't they just render everything to a local virtual terminal and then send the screen after updates?
How did Lotus 1-2-3/M address this? From what I understand, they adopted an approach commonly used in 3270 applications – a fixed area at the bottom of the screen was reserved as a "Command area". You would type your slash sequence there, and then press ENTER. The whole sequence would be sent to the host. If what you typed was not a full command, just the prefix of a command, I believe it would display you a menu showing the command options, similar to how 1-2-3 worked on other platforms.
But it also was a problem with things like scrolling off the screen. On 1-2-3 on other platforms, you could just use the cursor keys, and the screen would scroll immediately. Under 3270, the cursor keys are implemented locally, so you'd have to use a function key or command to scroll instead.
The big idea behind the 3270 architecture was to support things like CICS, where you could have tens of thousands of terminals attached to a single processor computer with 16 megabytes of memory, and then take individual requests from terminals, recover state, and render screens.