The same algorithm which is used by every other RAW editor on the face of the earth, including the now 15 years old Capture NX2 which had NONE of these visual rendering problems at any scaling level. FIFTEEN years ago, with much less powerful PCs, and no assistance from GPUs.
If it could be done then with no performance issues, it definitely can be done today too.
Previous discussion here, with PLENTY of examples:
From my point of view, it would be much better if corrections like CA, sharpness, aberrations were ALWAYS applied no matter the zoom level, so that the image is always completely rendered internally at 100% and THEN downscaled for display, even with a simple adjacent pixel averaging (as opposed to binning) algorithm. No visualization at less than 100% will ever be completely accurate, but it would already be a BIG improvement to what is implemented currently.