I have some suggestions to improve the manual (man zstd) and an error message.

After this manual section:

•   -T#, --threads=#: Compress using # working threads (default: 1). If # is 0, attempt to detect and use the number of physical CPU cores. In all cases, the nb of threads is capped to ZSTDMT_NBWORKERS_MAX==200. This modifier does nothing if zstd is compiled without multithread support.

I suggest adding this note:

Note that memory usage increases with each thread.

It might seem obvious to some, but still worth noting for those who don't know, to spare them the surprise.

Also, after the error message "Warning : compression level higher than max, reduced to 19", I suggest adding "Specify --ultra to raise the limit to 22 and use --long=31 for maximum compression. Note that this requires high amounts of memory, and the resulting data might be rejected by third-party decoders and is therefore only recommended for archival purposes." This explanation is taken from #435. I suggest mentioning long because someone who uses -22 probably intends to maximize the compression ratio, and they might not be aware that the --long option exists.

After

--long[=#]: enables long distance matching with # windowLog, if not # is not present it defaults to 27.

Please add "The highest possible value is 31."