There was also a big technical reason for that: graphics subsystems sucked back then. Hardware acceleration ranged from non-existent to primitive and RAM was tight. If you have two objects (window, control, etc.) displayed and they're rectangular, you can update one of them without needing to involve the other one at all. Since your entire display was a single buffer in memory, each application would draw directly in its chunk of memory. When you did something which made things move you could use simple array math to move the previously-painted block of memory to the new location or, if you had a really fancy graphics card, call a BitBlt operation which would move x1,y2-x2,y2 to the new location. Then you had to tell whatever was now visible in the source location to repaint that region.
Adding things like transparency or rounded corners breaks a lot of that — you need to know what's underneath to update the display — and that wasn't fast enough to use until you had enough RAM to store each object independently so you don't need to ask each application to repaint as things move around. Compositing window managers started early — I believe the Amiga had that — but the first mainstream OS was when Mac OS X shipped Quartz and it had performance drawbacks (i.e. tolerable but not great window dragging performance) until they shipped Quartz Extreme which used OpenGL to do the hard work. Microsoft caught up about 5 years later with Windows Vista, in part because they had more users to support with inadequate hardware.