"This is just a precision error - You will still get precision errors in floating point!"

Sure, but for most practical purposes the error is small enough to be irrelevant. There's a big difference between 2 and 2.5 though.

What I was trying to highlight is that if you use integer arithmetic to do stuff that is better suited to floating point you are going to have to think about it very carefully.

If speed is an issue it isn't difficult to calculate the worst case execution time for your floating point calculations (with Keil at least) to determine whether you can use it and save yourself a load of hassle.