That sure brings back memories. A friend of mine who is not exactly PC hardware-savvy got a 300A @ 450 built by a geek friend of his at college, and the OC lasted for 3 years. Once he was back from college, I had to diagnose and fix instability problems that started cropping up unexpectedly. All I did was drop the volts and lower the FSB and bam, the thing started working again. It was slow as balls at 300 mhz, but at least it would boot.
But as far as overclocking-as-a-business goes, I've long contemplated the subject, since I think it would be a fun way to do business even if the margins would not be great.
The main problems I see with "pre-testing" these days is really the same problem it has always had: what do you do with the bum chips? If you can milk some vendor's RMA system to cherry-pick the good chips and return the bad ones, then you have to mark up the good chips enough to offset the time and money it took to test all the chips and pay the applicable restocking fees. Clearly this is not quite as feasible as buying chips by the try and plunking the "bad" ones into OEM systems, but being a system builder these days . . . shew. That's a whole 'nother ball of wax.
Anyway, the end results of your efforts will be that a few people that wanted great chips would get what they wanted at a premium, and everyone else dealing with the retail channel would have to hope they didn't get one of your duds when buying OEM/open box. I think some people still do this, but I'd feel kinda oogy about returning chips like that when they worked as advertised.
Another problem I see is that, with so much of the system being crammed into the CPU package (cores, northbridge/uncore, memory controller, IGP), are you going to test only the maximum core speed for the CPU? Or are you going to test everything?
Then there's the question as to whether or not the buyer will even know what the heck to do with the chip once they've bought it, which is a scary prospect considering how volt-sensitive these chips are getting at 32nm. Imagine selling a pre-testing 980X to someone, only for them to pop the chip via excess VTT. What a mess that would be . . .
I'm sure there are ways around all that, but again, it'd be a lot of effort for very little reward and a lot of potential liability. You'd pretty much have to include all sorts of caveats with your pre-testing, such as specifying certified motherboard(s), RAM, and cooling. Then you'd have to specify voltage limits with whatever product guarantee you offered. Everything would have to be structured very carefully in the best legal-ese possible.
There are probably better ways to leverage overclocking-as-a-business than pre-testing, but I haven't quite sorted out what that might be.