In general, it's often worth doing transforms like RLE combined with general purpose compression. General compression algorithms don't know about the details of

In general, it's often worth doing transforms like RLE combined with general purpose compression. General compression algorithms don't know about the details of your data and typically have a max window size period, so if RLE compresses your data a lot, it makes LZMA (or most other compressors) will be seeing a giant block of zeros most of the time and won't be able to see nearly as much of the actual data. Running compression after RLE will mean that te giant chunks of zeros will be squashed down so the regular compressor can fit non-trivially compressable data within the window size and more usefully look for improvements.