With old and noisy sensors I found the best way to get a decent clean background was via dithering. However, while using my old Super Polaris the periodic error helped – though the defective pixels could cause a streaky pattern in the direction of my declination drift.
Realistically, using a series of darks to find the noisiest pixels and then spatially filter them might be worthwhile as the Gaussian random noise assumption made when applying median stacking may not apply well for CMOS sensors – each pixel has its own amplifier characteristics.