Problem:
A service station has three servers, indexed 1, 2, and 3 . When a customer arrives, he is assigned to the idle server with the lowest index. If all servers are busy, the customer goes away. The service times at server $i$ are independent and exponentially distributed with mean $1 / \mu_i$, for $i=1,2,3$, where $\mu_i \neq \mu_j$ if $i \neq > j$. The customers arrive according to a Poisson process with rate $\lambda$. Define a continuous-time Markov chain to analyze the number of customers at the service station and specify its parameters.
Part of the solution:
Answer Let $Y(t)$ be the number of customers at the service station at time $t$. However, the process $\{Y(t), t \geq 0\}$ cannot be directly analyzed by a continuous-time Markov chain. In fact the sojourn time for state 2 is not exponentially distributed. Instead we investigate $Y(t)$ indirectly by considering another process first.
What I do not understand yet is my specifically the sojourn time of state 2 is not exponentially distributed. It would be extremely helpful if somebody could show me the intermediate steps to arrive to this conclusion.