I did some work over Christmas to solve the near/far clipping plane calculation. But my plans were derailed somewhat due to Sega rally not having any valid node distances for the top level nodes, which made my far plane calculation not work. I've tried a whole bunch of things to fix this issue, the only thing I haven't tried is manually clipping the polys against the view frustum in software. I am pretty sure it would work, I'm just worried about the CPU hit. Hopefully the subset of polys will be small enough as to not be too expensive on the CPU.
But currently I am pretty short of time due to real life. When I have some time again I'll probably have another go in the future.
Thanks to Ville Linde I did identify some potential CPU bugs to do with floating point and division by zero. Really edge cases. Once a div by zero happens, it creates an infinity float value, which when multiplied by anything causes more infinities. This value ends up corrupting a good chunk of the float data in magical truck adventure in attract mode. I don't have a good enough grasp of the spec or the time to really look into this to fix it currently.