So it seems that the default changes automatically between the sound card or headphones: if I switch off the phones, continuous recording through the sound card of the PC automatically and vice-versa. That is why I thought (And still think) that was the safer option.
OBS will not follow the default device and change it on the fly. On OBS start or if you set some audio device to "default", OBS will determine the default device and use it until OBS terminates. This is different to common media players that change their output device on the fly, if they're outputting to the default device.
If your default device is a wireless headset, and it loses power mid-recording, OBS will continue to try to record it, even if it switches itself off. This will record silence. Because of this, you should not try to capture such volatile devices but instead choose devices that will not go away during recording.
About your initial question:
Audio is insignificant size compared to video, so focus on video and encoder settings to get a proper quality versus file size ratio.
Double the fps about doubles the file size.
Double the resolution increases file size about 4 times.
Encoder settings can vary the file size from nil (bloody mess) to insanely huge (as good as raw recording).
You didn't post a log and you didn't tell anything about the type of video you try to record (that matters!), so the encoder recommendation is generic:
For a long running video, choose advanced output mode and CQP/CRF/ICQ (depending on the encoder) as rate control for the recording encoder. Instead of bitrate, you're able to specify a quality parameter with these. It's called CQ level or CQP or CQP level or CRF.
This parameter is 0..50 and tells how much detail to remove. 0 = nothing, 50=all. Start with 25. A change by 3 about doubles/halves the file size. Make test recordings and choose the CQ/CRF value that produces the best quality/file size ratio for your type of recording. Before you do this, choose the lowest resolution and the lowest fps thats still acceptable to further shrink the amount of data to encode.