Actually I just did some little tests comparing mono and gcc on a inner loop with float math and one with int math, and it appears the mono/C# combo is actually 15-40%* faster on the float test, and only 0-5% slower on the int test.
So I think you can say it's as fast as native code (which is as expected because it's compiled to native code before execution...), if not faster. (because it can test processor capabilities and optimize code specifically for the host processor)
* depending on gcc optimization flags
