Is linear hashing good?

Considering the set H of all linear (or affine) transformations between two vector spaces over a finite field F, the ability of H as a class of hash functions is studied. Hashing a set S of size n into a range, having the same cardinality n by a randomly chosen function from H and looking at the size of the largest hash bucket, is particularly evaluated. If the finite field F has n elements, then there is a bad set S⊂F² of size n with expected minimal bucket size Ω(n^1/3). If n is a perfect square there is a worse set with largest bucket size always at least √n. If however, the considered is the field of two elements then better bounds will be obtained. The best previously known upper bound was O(2^{√log n}). This upper bound is reduced to θ(log n/log log n).