If you take a look at music like the two-part inventions and three-part symphonies by Bach, they make a lot of musical sense, and while they are intended as practice pieces, the prevalent problem (and increase of diffuculty when going from inventions to symphonies) is not one of hitting particular combinations of notes at the same time but rather of following several simultaneous parts. Even if we stipulate that listeners were expected to be sitting there score in hand (and at the price of good music copiers, I find that somewhat hard to believe: Bach had enough work writing down all the parts and delivered a lot just in time), the player would be expected to not distract in any manner from following the voices.
If you take a look at the first movement of Sonata III for violin solo (BWV 1006) in the original handwriting, it is strikingly obvious from the notation that even for a basically "monophonic" instrument like the violin, the player is expected in several passages to diligently distribute the voicing across two to three strings in a very particular manner even when crossing voices and echoing the same note in another voice. This same prelude has been turned into an organ prelude by Bach: the voicing remains even though the execution may be entirely different.
The characteristic distribution of voices across several strings is also prevalent in the lute version of this same prelude (the last few movements of the sonata sound a bit lacklustre on lute since they are actually monophonic): so the instrument again is expected to be played in a manner matching the voicing of the original.
Even if we are talking about instruments with fixed dynamics like a harpsichord, you will often arpeggiate simultaneous notes and can represent voice crossings by reversal of order. But larger harpsichords tend to have more than a single manual anyway, so properly distributing multiple voices is a lot easier than on a piano as long as the respective tessiture of ongoing voices don't force you to let a voice switch hands.
So if one sees how Bach lets the character of the "abstract" music surface on different instruments for the same piece, I don't find it easy to support the contentions that the voicing is just some theoretic thing in the score that the executioner is free not to bother with.
If Bach would have this view of his work, I'd have expected him to employ figured bass a lot more than he did since that pins down "just" the harmonic framework that results from playing several notes at a time.
But to put it in a nutshell: why would the common baroque cembalo go to the expense of providing two manuals if the keyboard player was not expected to bother with voicing?