Memtest/IntelBurn/Small FFT fine but blend test fails?

[Flaw]

I read the other topic regarding this, thought I'd still post another one as I saw this case being a tad different.
My system is a Q9550 OCd to 3.4GHz (FSB 400, 1.24Vcore, load temps under IntelBurn are around 65) on an Asus Commando with 4GB of Kingston HyperX 800MHz CL4 ram.
The PSU is a Zalman ZM600-HP, and my GPU is an Club3D Radeon HD4890 Superclocked Edition.

Overclocked:
I can run InterlBurn/Small FFT tests over night with 0 errors.
I've also just ran Memtest86+ over 20h with 0 errors.
Now, when I try the prime95 Blend test, it fails within the first minute.
Usually within the first 10 seconds, on all cores.

Stock settings: (by manually setting the FSB to the default of 333MHz, I don't trust the automatic values.)
Just as above except that Blend test now runs for max 5minutes on one of the cores, usually atleast one fails within the first minute.
I'm trying to troubleshoot the TDR issues I'm having with my GPU and came accross this rather peculiar behaviour with the Blend test.
Any ideas what could cause something like this?

Also, I'd like to point out that HW Monitor reports my PSUs +12V actually giving out around 13V, 12.97V-13.05V to be precise.
My multimeter is out of batteries atm so I can't confirm this value, but is it were to be correct, is something like that serious?
All other voltages reported by HW Monitor seem to be relatively close to what I know they should be.

Thanks for your time in advance.

[KingJames84]

ehhh I would get som batteries for that multimeter and test them rails. And while I was doing that I would be running the cpu @ stock along with the whole system and see if HW monitor changes to something that is closer together. Chances are your psu is fine but better safe than going oh ****...