Quake2 has a lot of different ways of benchmarking with it. There is the simple way of using 'demo1', which doesn't have much to do with game play reality, you can run Brett '3 Fingers' Jackob's 'massive1' and get some realistic average frame rate for multi-player gaming, or you can use Brett's 'crusher' demo, showing 3D performance under Quake2 at worst case conditions.
The latter is obviously the most important one, since we 3D gamers want to be sure that the frame rate never drops too low when we are having some heavy rocket fights with several opponents at the same time. Of all three different demos, 'crusher' is using most CPU resources, 'demo1' the least.
Comparing Rage 128 with TNT shows some interesting results. In 'demo1' Rage 128 runs 13% faster than TNT, this difference gets less running 'massive1' and comes down to only 3% in 'crusher'. These results are taken from a PII400 system. As you can see, the difference of Rage 128's Quake2 performance between the frame rates in 'demo1' and 'crusher' is much larger than in case of RIVA TNT. Rage 128 is going almost as low as TNT in 'crusher'.
These results do already lead to the conclusion that Rage 128 is more CPU depending than TNT. The mon2 demo from S3 is a very good test for the ability of a chip to handle large textures and shows the performance of the AGP transfer rate or texture compression, if available. The Rage 128 looks a lot worse than TNT, suggesting that the AGP interface needs a lot more work too.