Allowing for 2% bigger resulting file could mean huge speedup in these circumstances even with the same compression routines, seeing these benchmarks of zlib and zlib-ng for different compression levels:

https://github.com/zlib-ng/zlib-ng/discussions/871

IMO the fair comparison of the real speed improvement brought by a new program is only between the almost identical resulting compressed sizes.