Good points. I do satellite image analysis as part of my research, and most satellite data are used (panchromatic) at 1m resolution (these are often down-sampled to 2m to reduce the compute and storage load by a factor of four). To process the entire land surface of the world (150 Tpixel at 1m, give or take) in a week is quite a challenge, logistically and computationally. The new generation of satellites can give 30cm resolution, so roughly ten times more data: 1.5 Exapixel (ouch). Recognizing anybody at 30 cm resolution is impossible. Better resolution may be available in military satellites, but normally higher resolution work is done by aerial imaging.
What I do not understand in Google's reaction is why they do not apply some simple morphological filters to the image patch to remove the details. This is quite easy and fast. Using simple area-open-close, or levelling from markers you could remove the small features on the road without affecting the rest. Alternatively, edit out the data manually, and use image inpainting to stitch up the hole.