For case 1 and 2 , we can justify that the efficiency is T/T+S only since no context switch occurs because of timer interrupt which interrupts to preempt a process from running state to ready and then schedules another process.But the measured time for which the process runs given is T time units only.So we have to find efficiency keeping in mind this constraint of measurement.
Also for case 5 , we can justify the CPU efficiency will be 0 as time quantum approaches to 0.So useful time spent by any process on the CPU will also be 0 approximately.So this results in CPU efficiency of 0.
I hope this clears your query on different cases on round robin scheduling CPU efficiency based on varying time quanta.