Cyrix MII officially released models topped out at 300 MHz 2.2V model, though very rare, the 250, 262, 266 MHz models of 2.9 and 2.2V flavors were fairly common around 1998/1999 and the uncommon but not crazy rare 291 MHz PR 400. (they did take oddly long to embrace the relatively standard 100 Mhz Super Socket 7 bus speed, which also tended to have fewer problems than the 83 MHz speed or even 75 MHz, and easier to find proper motherboards to configure reliably, the 2.5x 100 model didn't come until 98 or 99, but a 2x100 model should've been on the table long before that; granted, you could set them to such speeds unofficially and they'd probably work ... though the older 6x86L or classic doesn't like 100 MHz FSB, I have several that run at 3x68 fine, but not 2x100, and yes 68, ie 204 MHz)

But many of the 2.9V models were made on 250 nm process and will undervolt a lot at stock speeds (I have a PR 366 2.5x100 model that's happy at 2.2V) while also taking overvolting like crazy (like 3.3V or possibly 3.5V without issue, though the WInchip2 is the same and also 250 nm, but node size doesn't really tell you tolerances alone given other manufacturing differences, like metal inteconnect layer and other things). AMD's 250 nm chips will instantly pop above 3.3V and often die if overclocked at 2.8V for long periods. (they tend to be OK up to 2.6V, but that's a pretty hard limit: it's often what was used to push K6-2 550s to 600 MHz) That 250 MHz rated "PR366" (which would be more realistically a 350 if compared to a K6-2 on general purpose office benchmarks) effortlessly runs at 300 MHz if actually set to the rated 2.9V ... I think I had it above that at some point for some benchmarks. (was playing around with lots of retro PC stuff about 15 years ago) OTOH certain other motherboards aren't nearly as happy with that same CPU, though I can't be sure with age/wear/tear vs just board or chipset quirks. (VIA chipsets are supposed to benchmark better with Cyrix parts, but the one I ran the most was an ALI Aladdin V based Asus P5A-B board)

Realistically, the ghetto build budget gaming tweak option for Socket 7 was pretty much ... get a Pentium MMX 233, run it at 2x100 on a Super 7 board, and possibly overvolt it to 3.3, maybe 3.5V and go for 300. (they rarely do more than 300, but do tolerate 3.5V ... they're Intel 350 nm parts, and the rare mobile 250 nm ones with 4x multipliers are weird and tend to not work right with anything but Intel TX chipsets, though there's also a very rare 350 nm 2.2 or 2.5V rated mobile 233 MHz version that probably overclocks well). In any case, good for software rendered quake, good for 3D accelerator FPU requirements, pretty good at MMX performance (K6, or at least K6-2 is better at same clock speed: K6-2 improved MMX execution on top of adding 3DNOW!), though MMX only matters for Unreal's software renderer ... so the Pentium MMX at 300 MHz is probably going to be better with a Voodoo card too. (maybe even next to a K6-2/500 or K6-III/450, though 3DNOW! drivers for Direct3D would probably change that, I forget if any 3DNOW! patches for Glide or miniGL wrappers were any good at all, I know the one for Quake didn't really work well enough to fully close the gap, well ... maybe for 500 vs 300 Mhz, but still)

I will say that playing Unreal with a Voodoo 3 at 1024x768 on a K6-2 300, and Cyrix MII and Pentium MMX both overclocked to 3x100 MHz, the difference in average and minimum framerate (without using a frame counter) was not immediately obvious, or at least not dramatic. (software rendering speed with all at same clock and bus speed also wasn't that dramatic)

Heh, you know Quake II ALMOST got MMX optimized, but Carmack opted out of that and focused on Pentium Pro optimized code instead (which still ran well on the P5 Pentium family), but had he gone that direction, it would've had bigger implications for Cyrix and AMD, albeit more for casual gamers not willing to shell out for a 3D accelerator. (or specifically, any one with useable OpenGL drivers, even beta ones ... obviously 3D FX was the best at the time, but I recall some of the beta drivers available for ATi Rage Pro actually did relatively better than the Nvidia Riva, though the latter was usually faster in DirectX; the Ati Rage forums back in the day and ATi's webstite had a lot more obscure tweak happy stuff that's either gone or really hard to track down these days: they had beta DVD media player drivers too at one point that worked quite well on our K6-2 300 rig, but probably nearly impossible to find anywhere now, and even then, my Dad had to patch together bits of several different versions of the software to make it work reliably ... and that lasted until we upgraded from Windows 98SE to 2000)