So, my instinct were half right. This is with direct transfer on a decently large payload. It spends the majority of the time encoding the data, and it hangs up the tui.

But this isn’t a fair test, there’s the shared mem approach, which would be the ideal marker