Recently, a new variant of networks of evolutionary processors has been introduced by Erzsébet Csuhaj-Varjú and Pramod Kumar Sethy in the Journal of Membrane Computing (‘Length-driven communication in networks of evolutionary processors’). In this variant, the filtering of words before communication is based on whether their length in the current node increases, decreases, or remains unchanged in the preceding evolutionary step. It was shown there that any recursively enumerable language is generated by a hybrid network of evolutionary processors with this new communication mode. The network constructed has eight nodes and simulates a phrase structure grammar in Penttonen normal form. In the present paper, we improve this result by showing that seven nodes are sufficient for generating any recursively enumerable language by such a kind of networks.