I've been following this for more than a decade, even wrote a lot of audio software for Linux, and all my music is made under it, with my own apps. Yet I recognize the situation will never improve. Here's why:
1) While the Linux kernel is perfectly capable of low latency, even on the shittiest of hardware, it does not provide the concept of primary and secondary buffers. If you want to use pro audio, you want to be able to mix the low latency, high sampling rate stream together with the regular OS/Desktop audio. Windows and OSX do this by setting the hardware for the realtime client, then also mixing the secondary audio over it, which comes from userland (or already mixed in userland). As a result, when using realtime audio in Linux, desktop audio dies or is hacked to route pulseaudio to jack and other stuff that does not really work well.
2) It's impossible to write plugins similar to VST, because of the different way tookits connect to X11 (they won't share the connection). You can't mix and match toolkits so a host DAW will use different plugns. The only way is to use separate processes, but that makes programming complexity much higher and very few people bothered. Wayland seemed like it could fix this in the future, but other distros such as Ubuntu refuse to use it, so it doesn't seem good.
3) Good programmers are not necesarily good composers. This is something that is much more important than it seems. Commercial companies are forced to listen to their users, but OSS developers mostly care about doing something good enough for themselves. Given the chance that a good programmer is a good producer/composer is super slim for the practical world, most audio software kind of sucks and feels incomplete. Ardour took more than a decade to implement MIDI and it still is horrible, because the main developers care more about live session recording. If they really had to use it everyday to make professional music, it wouldn't be as bare bones as it is now. At the same time, stuff that looks like a good idea (jack daemon) are terrible in practice because making music with a bunch of applications open is akin or worse to a live set of devices with cables connected.
4) Finally, the biggest problem of Linux is that, unlike other software such as 3D or imaging, there is plenty of cheap and good Windows/OSX audio alternatives, so even if OSS software were to run properly on Windows/Mac, the incentive is still slow. It's not like Blender or Gimp, that it's commercial counterparts are in the thousands $.