Compression of short text strings, such as the GSM Short Message Service (SMS) and Twitter messages, has received relatively little attention compared to the compression of longer texts. This is not surprising given that for typical cellular and internet-based networks, the cost of compression probably outweighs the cost of delivering uncompressed messages. However, this is not necessarily true in the case where the cost of data transport is high, for example, where satellite back-haul is involved, or on bandwidth-starved mobile mesh networks, such as the mesh networks for disaster relief, rural, remote and developing contexts envisaged by the Serval Project [1-4]. This motivated the development of a state-of-art text compression algorithm that could be used to compress mesh-based short-message traffic, culminating in the development of the stats3 SMS compression scheme described in this paper. Stats3 uses word frequency and 3rd-order letter statistics embodied in a pre-constructed dictionary to affect lossless compression of short text messages. This scheme shows that our scheme compressing text messages typically reduces messages to less than half of their original size, and in so doing substantially outperforms all public SMS compression systems, while also matching or exceeding the marketing claims of the commercial options known to the authors. We also outline approaches for future work that has the potential to further improve the performance and practical utility of stats3.