I'm writting a GUI system in DX. Currently, I'm using win32 GDI calls like CreateSolidBrush() and CreateRectRgn() to draw my windows and controls to the back buffer. The win32 API's require 24bpp colour values while I'm using a 16bpp DX mode. Should I be writting my own drawing routines for drawing boxes and such? Which way would be faster? Which way is used more?

It would probably be much more efficient to draw your own GUI system via DX. Check out the articles here on GameDev. There was a series that might give you some good ideas.