While it is interesting that a single texture may leverage several threads/several cores to encode with etc2comp, we must acknowledge that a typical encoding jobs runs on dozens, hundreds, thousands of textures. These jobs are “embarrassingly paralellizable” (because each texture is processed independently), so any single-threaded algorithm (Mali?) would in practice be paralellized with minimal effort.

To be fair, we must compare performance of either (single-threaded-etc2comp VS single-threaded-Mali), or (multi-threaded-etc2comp VS multi-threaded-Mali).