The limitations of 1366x768 only became obvious in denser Liszt music.  Only by previously knowing what it should sound like is it legible.  The original PDF isn't very legible either.  Maybe there's a smarter way to preserve more details than thresholding & using RGB channels as pixels.  The x dimension is what kills it.  There would have to be a translation of common symbols in the PDF to a library of vector graphics, then precisely positioning & scaling the vector graphics.  Quite an effort with little gain in an age of 4k panels.  If the input pixels alternate horizontally between light & dark between the output pixels, it could force the output pixels to alternate horizontally without regard for position.

Upgrading to 1920x1080 panels is tempting & would allow everything to run on a single raspberry.  16x9 is not as ideal as 4x3, but there aren't any 4x3 panels.  It would also entail ordering 2 sets of LVD cables.