The output of any scanner is an image/picture of whatever was on the paper feed into the scanner. If the paper had text than the image is of text. Within a range of parameters an image of text can be processed through OCR to obtain an output of recognized characters. The output is character by character - there is no word processing like development to this. OCR supports find/search.
As to the scanner's image. An image can be created to have a small or to have a large footprint. The smaller the file size the less acuity / detail / quality there is for the image. An image that is created with an adequate resolution (if of text the rule of thumb is always have no less than 300 ppi (dpi is actually for the printer output).
Anyway - to compress an image the software has to start removing parts of the image.
(Lossy compression is the destructive removal and is at the heart of JPEG compression - JPEG 2000 attempts to mitigate this).
If you want good user quality avoid adaptive compression. Avoid any compression routine that use Lossy compression routines. For scanner output intended to be usable by humans "small" is Not better. As you have observed, the end product is -- let us say inadequate eh.
If a page of a post-processed PDF has rotated simply use Acrobat to rotate the page to the desired orientation.
Be well...