According to my tests, any extra communication across tracks (ie. send/receive, lh_midiinout or OSC) introduces the same latency bound to Live’s audio buffer. I think it’s because different tracks are running on different threads and switching between them (scheduling) is determined by the audio buffer size.
How did you implemented your tests? I would like to replicate them in my computer,
I did two simple devices, but i didn’t detect any latency.