Dropping one or two observations does not spoil consistency

Let $X_1,\dots,X_n$ be a random sample from exponential distribution with scale parameter $\theta$. Define $T_1=\bar X_n$; $T_2$: mean of the first $(n-1)$ observations; $T_3$: mean of the $(n-2)$ observations excluding the first and the last. Then: