Spaces of matrices with few eigenvalues

Let K be a (commutative) field with characteristic not 2, and V be a linear subspace of n by n matrices that have at most two eigenvalues in K (respectively, at most one non-zero eigenvalue in K). We prove that the dimension of V is less than or equal to n(n-1)/2 provided that n be greater than 2 (respectively, the dimension of V is less than or equal to n(n-1)/2). We also classify, up to similarity, the linear subspaces of n by n matrices in which every matrix has at most two eigenvalues (respectively, at most one non-zero eigenvalue) in an algebraic closure of K and which have the maximal dimension among such spaces.