For that you can use https://github.com/dimtpap/obs-pipewire-audio-capture
Add a new souce of this type, and it will add a new audio track, for which you can select which application(s) you want to be captured for that source. Works pretty much like the pipewire screen capture source in obs, but for audio.
I usually have 3 of these sources running: Discord, Game and Music player audio; in addition to a normal full desktop audio capture and my mic. They all go on separate audio tracks in the output file.
Some filesystems (like btrfs and ZFS) support comression on the filesystem level, where each block is compressed with some algorithm automatically, completely transparently to applications.
Most modern CPUs are fast enough at the light levels of compression that is used, that usually its also faster, because you read less data, and the read + decompress time is often lower than how long it would have taken to just read more data; though of course that depends on what data exactly, but overal its often faster (though usually its not by a very significant amount) for most average uses.