I duplicated your behaviour (Turning on T1 with T2 already on also turns on T3) with the Labs->Virtual Switch implementation.
However, if you use one of the community virtual switch implementations (I tested with TAustin’s) then you get what behaviour you expected to get. If T1 is off, T2 is on and T3 is off, and you turn on T1, T3 does NOT turn on. This is what I assumed would have happened as well.
This appears to be that the built in virtual switches have different rules to play by. Normally if an Edge device is already in a state and its requested the device change into that same state, no rules or logic run. Labs Virtual Switches do not appear to have to conform to that rule. If a Labs virtual switch is already in its state and its requested to go into the same state, rules will run again.
Very interesting!