It's all to do with the shock waves generated just ahead of the bullet tip which fan out like an inverted V with the bullet inside.
At very high supersonic speeds, the shock waves have a greater amount of separation from the bullet reducing turbulence and drag along its edges. As speeds drop, the amount reduces, turbulence around the bullet and drag increase peaking just above the speed of sound. Transonic speeds are a related but separate issue that sees a large increase in bullet drag and turbulence around it as some parts of the airflow around the bullet are supersonic and others sub as the air follows the bullet body's shape. Once on the speed of sound, the supersonic shock wave disappears entirely, likewise most turbulence around the bullet and drag is very much less as a result.
That's why the typical drag curve as seen for example in Bryan Litz's / Applied Ballistics books shows a curve that starts low (at the right which is the high velocity side), rises steadily as you go left with reducing velocity, more steeply through the transonic zone and peaks just above the speed of sound with an even faster rapid fall to the left (subsonic) side of the 1.0 MACH line.
http://bulletin.accurateshooter.com/?s=bullet+drag+&submit=Search
That's my layman's understanding and there are now doubt many wrinkles and subtleties in the precise nature of what affects drag at different speeds. For instance, we now know that a higher degree of gyroscopic stabilisation is needed with Sg values of 1.5 and above to reduce various flight behaviours that produce extra drag with modern very long for calibre streamlined bullets.