P95 27.7x does have AVX instructions, just not the newest AVX. I think you are wise to avoid the newest version of P95 for as you already know it skyrockets temps about 10c higher than the 27.7x version. Personally, I think the older version of P95 is just fine for establishing stability as long as you can pass it for 2 hr. or so. Driving temps through the roof doesn't prove anything about stability necessarily. When running normal apps you won't get anywhere near the temps that P95 27.7x produces, much less those produced by the latest Prime95 version. I'm using P95 27.9 myself. But I also use a other apps in a cocktail to get a more well-rounded stress test. So I use P95 and also IBT (high/very high setting), AIDA64 and 3DMark. P95 is still the main tool in the arsenal however.
I recently got a G3258 and was disappointed. I could not get it past 4.3 on 1.275 vcore. Adding more vcore doesn't seem to help, at least to the point where I would feel safe running it 24/7. It seems that many of the earliest production G3258s would commonly get to 4.8 ghz or so. But as production of the quad core Haswell's became more reliable and fewer duds were turned out and made into G3258s then those high clocking G3258s were no longer available. At least that's the explanation I ran across and it makes sense to me as we have seen this phenomenon before.
You might also mess with the Ring ratio and Ring voltage. A higher Ring Ratio improves performance because it speeds up the cache. Try about 33x and 1.2v.
The research I have done makes me wary of going higher than 1.3 vcore on the Haswell CPUs for 24/7 use, even if temps are controlled.