First, it's important to understand that MPC is not possible for $t \geq n/2$ in the information-theoretic setting (where the adversary is computationally unbounded). In the computational setting, where the adversary is polynomial-time, it is possible to achieve MPC under cryptographic hardness assumptions.
Regarding your specific question, I assume you are referring to a case where we have a polynomial of degree $t$ to share a secret. Indeed, it is possible to share a secret securely for any $t\leq n$. However, the problem is that you need to be able to process that secret. So, if $t=3$ and $n=6$ and you have a degree-3 polynomial, then the first step of multiplication yields a polynomial of degree-6. In order to reconstruct the values, 7 points are needed for a degree-6 polynomial. However, there are only 6 parties and only 6 points. So, this cannot be done.
Of course, the above just says why this approach doesn't work. However, it's possible to prove that no approach can work, and I believe that this also appears in the book.