Do you use a one to one correspondence for viewport pixel to object on the screen? I just wrote a blur function today in Blitz that takes the color information to pixels adjucent to the current pixel and averages them out. Then the whole picture is blured that way.

In a 2d Blitz application this process takes about 1.5 seconds. I'm wondering how fast you're able to implement this using a vector graphic like buffer?

P.S. Blurring a picture isn't really heavy and abstract math but I don't expect you to give out your secrets if it really bothers you.