Practically any 120mm tower heatsink will be enough to get you into decent OC territory, you just won't be able to push the voltage as high, so that last 100-200MHz require an expensive cooler.
On the P67 (I didn't have the Sabertooth, I was running the P8-P67M B3) it was trivial to overclock the 2500K:
- Leave the voltage alone and change the turbo multiplier to 40 (good starting point)
- If it boots to windows, restart it and increment the multiplier by one.
- At some point it'll crash booting windows - at stock voltage that is usually somewhere between 4.2 and 4.6GHz.
- Leave the multiplier where it was but add 50mV to the vCore and see if that boots. If not, drop the multiplier by one.
The aim is to see how far you can push it at stock voltage and if it's on the verge of being stable at any given speed, a minor voltage bump to ~1.25V should be enough to keep it stable without pushing the power requirements up too far. 1.35V is the realistic limit of air cooling and 1.4V+ is definitely either deafening or into watercooling territory. I've seen a 2500K that couldn't hit 4.2 on stock, and I've owned a 2500K that could hit 4.7 on stock. The key to low temperatures is to keep the voltage in check. 75mV is about all you want to add otherwise your temps will skyrocket quickly. It's an exponential power consumption curve as you increase the voltage!
Cryorig, Noctua, Coolermaster - anything 4+ heatpipes, 120mm+ and decent manufacturing quality should be able to dissipate 125W or so, and that's what you'd expect from a 1.25V clock in the 4.6-4.8 range if your chip can get there. Whether it gets there is just luck of the draw though and if you want to go that high there's a chance that will be from a leaky chip sucking down 1.4V or more - which is watercooling territory I'm afraid since you're looking at 165W or so (I'm estimating the power consumption of a email@example.comV based on 95W at stock).