I recently picked up an old Radeon 7000 for testing purposes and found it exhibits depth fighting/clipping problems, especially with multi-pass surfaces. After a bit of tinkering with the depth buffer and transform hints I finally discovered compiled vertex arrays were to blame...

The 7000 doesn't have hardware transform, so compiled arrays aren't going to help it much anyway, but I didn't expect them to break things either. Now, I can work around this easily enough but I was wondering if I can expect the same problem on other cards that lack hardware transforms - e.g. Rage 128's - I think all other Mac OS X compatible cards have hardware T&L and should be OK, but any insight would be appreciated.

Are you entirely sure that you are not modifying the data between Lock and Unlock calls? If so, you should write a small test case and file a bug with Apple.

Otherwise, I _very strongly suggest_ you forget that Compiled Vertex Arrays ever existed. Vertex buffer objects are superior in every respect, and are, by far, the most future proof-way to submit vertices.

VBOs are supported on every OpenGL renderer on OS X 10.4, including those without HW TCL.