For dense graph $E=O(V^2)$
Prim's Algorithm with Min heap takes $O((E+V)\log V)$ time
Array implementation takes $O(V^2)$
So, for dense graph
Min heap takes $O((V^2+V)\log V) = O(V^2 \log V)$ which is asymptotically more than $O(V^2)$
For sparse graphs, $E = O(V)$ and the time complexities for binary heap and array implementations will be $O(V \log V)$ and $O(V^2)$ respectively.
So, heap implementation is suitable for sparse graphs and array implementation for dense graphs.