Photoshop's blur is not realtime. Actually, it usually takes more than one second to calculate, I'm sure you don't want that kind of framerate.
The reason for photoshop's sluggishness is that it calculates a proper blur with the amount of samples equal to the size of your blur. You can't do that in realtime.
HOWEVER, the gaussian kernel is separable, which means you can calculate a new blur with the results of the previous and the resould would be the same as taking b^n, where b is the number of samples of the original effect (12 in this case) and n is the number of times you execute the effect. The real size of the final blur will be s*n, where s is the size of the blur you're stacking, asuming they're all the same.
At this point I feel it's a good idea to point out that ALL AAA 3D games do more than one pass of blur. So this is what you want to do, just like they do it.
SO, for a pixel-perfect blur, you'd have to use a size of 6 because that means one sample per pixel (6 to each side, total 12 pixels for 12 samples) and then stack as many blurs as you need. Say you want to blur to about 100 pixels as in photoshop. That'd mean s*n = 6*n = 50
then n= 100/6 = 8.3, you need to stack blur about 8 times on each axis. yeah.
And yeah, that blur is per-pixel accurate up to size 6.
Then again... I did say per-pixel accuracy is not feasible in real time. Let's use a blur that's a quarter as accurate, so size = 6*4 = 24. Now 24*n=50 means n=2.0
Stacking blur two times for each axis will yield a quarter of the accuracy of a per-pixel blur for a 100px wide blur. Inefficient? Just like Bioshock and COD4 (only they use a trick we can't do in Construct to use even less samples, meaning they're even less accurate).
Then again, if you attempt to stack ANY effect more than twice, CONSTRUCT CRASHES.
Yeah. Did report it back then, still happens in 995. Wish it didn't.
All that said, I've never tried blurring with blending. Can I get screenies of what goes wrong?