It's been a while since I used my Voodoo 5. Paired with a fast enough processor (I suspect ~1GHz+) it generally performs somewhere between the Geforce 256 SDR and Geforce 256 DDR, so think GeForce 2 MX, though it can outperform the 256 DDR depending on the scenario. This will depend on the resolution and game used however, generally speaking the Voodoo5 is weaker in low resolutions, it tends to flex its muscles at 1024x768 and above.
Here's an interesting comparison with mostly later games.
FSAA 4X is nice, but it's only really usable for games that are locked at 640x480, so especially useful for older Glide games. FSAA 2x is not that useful IMO.
In general, you can expect half the performance for every step of FSAA applied. So, in terms of performance, the following applies, roughly:
2x FSAA Voodoo 5 ~= Voodoo 4 (V5 is quite literally 2x Voodoo 4 VSA100 chips!) ~= Voodoo 3 3000-3500 ~= Voodoo 2 SLI
4x FSAA Voodoo 5 ~= 1x Voodoo 2
I hope the above is readable, basically at 4x FSAA the Voodoo 5 performs roughly on par with a Voodoo2, though obviously without the texture thrashing issues of the Voodoo 2 which has split memory pools for framebuffer and texture memory. So it is fine for 640x480, less so for 800x600 and above.
Pushing negative LOD bias is indeed kinda like fake anisotropic, it's actually quite nice. Of the games you mentioned, the Unreal Engine based games will run better on the Voodoo 5 than any other card of that era and generally speaking more "correct", as the D3D renderer on these games was a bit dodgy at times. Hope this helps.
