A) depends on workload, K6-2+ is a very different beast to Intel's CPUs of the time.
K6-2+ had 3DNow, which was very, very fast in the very few things that supported it. It had a typically weak FPU (at least when compared to the P6 monster), and again compared to P6 was slightly stronger in 16b code and weaker in 32b code.
So:
- if your game actually uses 3DNow, it will run better on K6-2+ than a similarly clocked P2.
- if you're running FPU-heavy stuff (Quake...), a P2 will get 50% higher performance at the same clock, i.e. perform equivalently to a 50% higher clocked K6-2+. So at best you CPU will be comparable to a P2-300.
- if you're running ALU-heavy 32b stuff (most other mid 1990s games), the difference will be at most 10%, so comparable to a P2-400 at least.
- if you're running ALU 16b stuff (i.e. DOS games), the K6-2+ can beat a P2 clock-for-clock - but will suffer from slower bus speed, but parity with P2-450 would be possible in Doom, I suspect.
You say your overclocked AGP port is the issue stopping you running at 83MHz. Maybe it's worth trying some different AGP cards. I had it running at 83MHz with AGP. Alternative is to go PCI, as with the ancient AGP implementation on the SiS 5591 it's not really giving you significantly more performance.
B) generally the best voltage to set is the lowest that leaves the system fully stable. The K6-2+ CPUs were intended as laptop parts and designed for low power. You should be able to get away with a significant undervolt at this level of underclocking, but just one way to find out with your CPU: try it, stress it and if it's stable, keep it.