Multiple sequence alignment (MSA) and phylogenetic tree reconstruction are one of the most important problems in the computational biology. While both these problems are of great practical significance, in most cases they are very computationally demanding. In this paper we propose a new approach to the MSA problem which simultaneously infers an underlying phylogenetic tree. To process large data sets we provide parallel implementation of our method, which is based on the distributed caching of intermediate results. Finally, we show a parallel server designed for grid environments, and we report results of experiments performed with actual biological data, e.g. 1000 ribosomal RNA sequences.