Sure, we could so that. As you can see xz compresses still better than zstd on high levels, but the difference to xz goes down from 18 to 7MB, or 28% to 12%, in your example. Not sure if that would be enough for Dominique. My experiments showed that at levels >~15, compression with zstd may actually be slower than xz (that was with initramfs payloads). *decrompression* with zstd is always at 10-15x faster than with xz, independent of the compression level. Also, multithreading can be enabled with zstd, speeding it up (can't be done with xz easily, because compression results may not be reproducible any more).