Because the die-heatspreader interface is often terrible on IB (do a search, tons of pages on it, IB uses paste for the interface as opposed to solder used in SB). But how terrible is variable. CPUs die when they short out. When the current is too much for the 'correct' pathways to accommodate, the current will find an alternate path (arcing). Resistance is the limiting factor, and temperature is a good surrogate for resistance. Liquid nitrogen allows such high overclocks because at lower temperatures, the circuitry can handle more current.
Let me put it another way, which would you be happier with, 4.5Ghz on 1.25v and temps in the 90's, or your set up? Obviously yours is better. So, do some searches to see what kinds of temps (ignore voltage) people have seen at certain levels of over clocking using a CM212. I doubt you will find any better than your result.
There is always a lottery when getting a chip. Prior to IB, the lottery only related to the die itself, how many volts it needed for a certain OC. With IB they added another variable, how well the die-heatspreader interface ended up. It seems you got a poor result in the first lottery, but a great result in the second. If you exchange your chip you will most likely get a better result in the first, and a worse result in the second. If your main goal is lower voltage, by all means exchange away. If your main goal is actual results, I'd think long and hard about it.