Interesting question.
I think Nick's answer conveys a lot that isn't generally realised: often limits are not real limits per se, simply the maximum demonstrated during testing.
As a real example of the point in question, the EC-155 has a maximum approach gradient of 4 degrees. What is the (unusually steep) ILS Glideslope at Marignane (Marseille/Provence)? 4 degrees.
Out of interest Nick, how much easier is it to do the tests prior to certification from a paperwork point of view, as compared to amending certificated limits later? Any difference?