First of all, glFrustum goes into the PROJECTION matrix, and unless you need to overlay something, you set that once (or once per frame) and then switch to the MODELVIEW matrix, which is where you translate, rotate and scale your world and objects. They're explained in the OpenGL red book and blue book (look for glMatrixMode).
Then; if you don't want perspective projection, then don't call glFrustum, butt use glOrtho instead. glOrtho basically keeps x and y coordinates intact, and uses z coordinates for depth buffering (and nothing else). This is in fact an isometric projection, and if you ignore depth buffering, you have in fact a 2D API. glOrtho, too, is described in redbook & bluebook.
There's one caveat to glFrustum and glOrtho: they only work properly if you first glLoadIdentity into the projection matrix, and then call either function respectively - otherwise the (unknown / undefined) former contents of the projection matrix are multiplied by the frustum or ortho projection matrix, and trust me, that is not what you want.
Third; read up on glPushMatrix() and glPopMatrix(); they allow you to save the current matrix status and restore it later. Very useful, and the proper way of doing things. With a simple scene, you'll need to call glLoadIdentity no more than once per frame.

So, in short, rendering a frame goes like so:
1) switch into projection matrix mode
2) load identity
3) glFrustum or glOrtho
4) switch into modelview matrix mode
5) load identity
6) apply camera transformation(s)
7) for each object, push matrix, apply model transforms, render, pop matrix
8) flip

Thanks a lot! , things look MUCH better now.

It doesn't depend on driver or at least shouldn't if driver wants to comply with OpenGL specification. You can even remove step 2 as all matrices are identity matrices by default and you can do step 4 just once.