If the G400 and G450 have the same architecture and both utilize the same OpenGL ICD, why is the code path for the Matrox G450 less buggy than the Matrox G400?

If the problem is entirely software related, I am curious as to whether it's possible to fix this by patching the codepath for the G400 to use the one in the ICD for the G450.

The issue stems from the G400 using a 16-bit z buffer in OpenGL when a higher bit buffer is requested (24 or 32-bit?). This is a known issue that was never fixed in the final beta driver release for Windows 9x (v6.83.017). Note that I am also using the latest OpenGL ICD, which is actually found in the 2k/XP driver (v5.96.004.0). The latest ICD is version 2.60.072.

I also tested using the AGP G400 vs the PCI G450, but I don't think that this would make any difference with regards to this specific bug.

Video demos can be found here:

https://www.youtube.com/watch?v=x3QmoCvlAVw
https://www.youtube.com/watch?v=-hGiRLjrGg0

I would prefer to use the G400 over the G450, since it's faster in games, but I also want to play Quake with OpenGL.