When the MRC compression algorithm seperates the text from the background image, you can still see the text in white there. Then, the background image is compressed using an ordinary image compression algorithm similar to JPEG. If you use JPEG to compress any image with textual content, you get this kind of halo as a compression artefact. You can often see this (try it out!) when people make screenshots and then save them as jpg instead of png: The JPEG compression artifacts are often so strong that you can barely read the text on the screenshot anymore.
I do it daily. Work is processing graphics and photos (then color separating) for printing.