It’s been years since I stopped using Preview for PDF files, generally using Skim for reading and pdfpen for editing them. Perhaps something changed since then? That is good news, if so.
I should have tried it before replying, but did not have easy access to my laptop. Now I redid the experiment. Starting with a 24 page scanned pdf, I deleted the first page using Preview, then saved and closed the file. It swelled from 2 MB to 6.7 MB. Looking into the file, I see that all the images in the original are encoded with CCITTFaxDecode, while the images in the edited file are encoded using FlateDecode. I cannot see anything relevant to the issue in the preferences.