In cycle stealing mode transfer takes place word by word..
Here word size given = 4 B
Here time CPU busy which is also known as preparation time = Data size / I/O Device transfer rate
In 1 s , data transferred = 104 B
So time take for 4 B = 10-4 s * 4
= 400 microsec
And time CPU is blocked ( also known as transfer time ) = 50 microsec
So % time CPU is blocked = (50 / 400) * 100
= 12.50 % [In cycle stealing mode , pipelining is there]
So by this amount only CPU activity is reduced..
Hence 12.50 % should be the correct answer..