Next, one thing I'd like to clarify - when General Track Info -> Output has two lines, the one in parentheses is what XMPlay decodes the file as, and the one without is what gets sent to the driver?
And well, the main confusion...
Your card supports up to 24-bit audio. In exclusive mode, XMPlay directly controls the audio card. If the file is 16-bit, XMPlay will output it as 16-bit. If you tell the WASAPI plugin to "Use highest available resolution", XMPlay uses the 24-bit of your card, regardless of the output resolution you chose on the output tab. Your card does not support 32-bit, so XMPlay automatically falls back to 24-bit if you request 32-bit.
So, do I understand right that in shared mode, all audio is resampled to the Default Format value at 32bit precision, and then downscaled to the output resolution?
Yes. In shared mode, Windows handles the audio mixing of different applications, so XMPlay feeds it with the highest quality audio data.
And if so, should I set it down to 44100 as majority of my music collection is 44100?
That's what I did as well. Works like a charm (though I don't hear the difference between 44.1 and 48 kHz
Would it be better to downmix 5.1 to 2.0 in WinampAC3 plugin's panel, as I did on XP, or better to let it emit all 6 channels and let Windows downmix it?
That doesn't really matter, unless you want some specific downmixing matrices, in that case you should use WinampAC3's config.