It's slow because of the multiple times it has to compile and optimize for all the different processor revisions.

Does it? Last I looked it did 1 by default, 2 if you ask nicely.

The main reason it's slow is that it does a lot of analysis, even at -O1.

It was my understanding that the advantage of Intel's compiler is that it would optimize for the strengths available on newer processors while still allowing it to work on older processors (for example, using AVX instructions if available and a slower branch if not) [aka, the "CPU dispatcher"]. Agner wrote about it "crippling" AMD processors because they didn't say they were "GenuineIntel" back in 2009.[0]

[0]: http://www.agner.org/optimize/blog/read.php?i=49