As stated by the man/help page, these levels require more memory.
This is an understatement.
Most importantly, the increased memory requirement will impact not only the compression side, but also the (future) decompression side.

To require a lot of memory on the decompression side is not welcomed. It could be considered an attack, trying to take away memory from the receiving system.
For its own protection, a decoder is free to reject a frame which requires too much memory.
So, in short, these frames are more likely to be rejected by 3rd party decoders.

As long as you remain in the x64 PC world, using the official zstd cli, there will be no such problem. Here, memory is plentyful.
Problems will start to happen when trying to share such frames with lighter systems, such as, for example, a mobile 32-bits ARM platform. These platforms will typically reject frames built with levels 21 and 22. It can be even worse, as systems are allowed to set their own limit.

In order to reduce the number of misfits, a memory threshold is recommended, keeping most systems compatible with each other. This value is 8 MB. It's used by level 19.

So basically, when setting --ultra, it makes the compressor go beyond the threshold, and reduce the number of systems which will be able to decode the compressed frame. It's a non-issue as long as the frame will be consumed by some laptop/desktop/server class, typically x64 Intel chip platform.

If you like the setting, and are tired of having to type it manually every time, you can alias the command, so that it will be enabled all the time (and it isn't a problem for faster modes) :
alias zstd='zstd --ultra'

Hope it answered your question. ping me it not