Yes, I'm aware of this problem. It does not occur on any of our Macs with X1600 cards, but they all run Leopard, not Tiger. So I'm afraid my fix for the death screens triggers a bug in the Tiger drivers. That's not entirely unexpected, considering that the fix uses a feature that is only partially supported by the hardware, but advertised by the driver.
I'll do a test with Tiger tomorrow and see if I can reproduce the problem.
For the technical background for those that are interested:
Radeon cards up to the X1000 series have a limitation which limits them to textures that have power of two sizes: 1, 2, 4, 8, 16, 32, 64, 128, ... . Nvidia cards from geforce 6 and newer do not have that limitation, but earlier Nvidia cards have it too.
Now ATI cards have conditional support for non power of two textures(e.g. a size like 100x100). This conditional support has certain limitations, for example no mipmaps, and certain texture address modes are prohibited. Team Fortress 2 uses such conditional non power of two textures in the death screen.
Textures in Direct3D and power of two textures in OpenGL use so-called normalized coordinates. Ie, the texture coordinates range from 0.0 to 1.0. So 0.5 specifies the middle of the texture. This applies for conditional non power of two textures as well in d3d.
Now there's an extension for conditional power of two textures in OpenGL: GL_ARB_texture_rectangle. It matches conditional non power of two support in D3D, except that it uses non-normalized coordinates from 0 to with and 0 to height. Ie, width/2 specifies the middle of the texture.
In 7.0 and earlier we used this extension for conditional NP2 textures, with a correction for the coordinates. However, this coordinate correction failed in Team Fortress 2, and it is not possible to get it universally working with shaders.
So now MacOS advertises unconditional NP2 support on radeon cards. Power of two textures created with this emulated NP2 support have normalized coordinates, like in D3D. This universally fixes the addressing problems, but when CrossOver outsteps the limits of conditional NP2 textures, the driver falls back to software.
So now I am using this emulated unconditional NP2 support, but I'm extra-careful not to step out of the conditional NP2 limitations, thus avoiding the software fallback. This works fine on Linux and Leopard, and reportedly fixes the incorrect death screen rendering.
However, apparently Tiger does not like it.