One underappreciated reason DEFLATE/gzip persists: it's streaming-friendly. You can emit compressed bytes incrementally without buffering the entire input. BWT-based algorithms (bzip2) need the whole block in memory before the transform can run.

This matters a lot for formats like ZIP, where the archive is constructed on-the-fly. You can pipe data through DEFLATE and compute CRC32 as you go, writing headers retroactively via Data Descriptors. Try that with bzip2 and you're stuck buffering 900KB blocks before you can output anything.

zstd threading and framing are a step forward here, but DEFLATE's simplicity and tiny memory footprint (~32KB window) make it hard to kill in constrained or streaming contexts.