gzip compression levels are almost useless and typically result in very little actual compression ratio differences, but typically with massive CPU usage differences.
zstd doesn't do everything better than gzip though; I believe gzip still offers somewhat faster compression for equal compression ratios last time I looked (but decompression is much faster). I mostly replaced gzip with zstd myself, but there are still scenarios where gzip might be preferable.
In all the testing I have been doing across a variety of use cases, from BigData, to databases, to games, I have not come across a scenario where Zstd is doing worse than Zstd.
Compression time CPU is lower for equivalent ratios. Maximum compression is better. Decompression CPU is much lower (i.e. faster), and that is independent of the level used at compression time.
I did a bunch of testing about five years ago and in some scenarios gzip was a bit faster, but I don't recall which exactly and it was five years ago so things may have changed. I thought I had saved the results somewhere, but I can't find them right now.
Overall, zstd still came out as the clear winner (also compared some other compression tools), so I went with that.
gzip compression levels are almost useless and typically result in very little actual compression ratio differences, but typically with massive CPU usage differences.
zstd doesn't do everything better than gzip though; I believe gzip still offers somewhat faster compression for equal compression ratios last time I looked (but decompression is much faster). I mostly replaced gzip with zstd myself, but there are still scenarios where gzip might be preferable.