I'm afraid that's the way it is. I faced the same problem some time ago. You may find tricks to estimate the inl or conversion error, but at the end of the day there's no way out. You need a number of samples per code and you need to simulate them, so, yes it takes ages.
The only thing you can do is to simulate conversion error instead of inl, that's just hitting once each code. That will give you a worst case limit for inl. For dnl is a nightmare too.
What I used to do is simulate 1 hit per code for the full range and then in the difficult parts (in my case, being pipeline was at 1/4 vref) simulate 20 hits per code.
The other thing that could be done is to simulate in parallel dividing the full range in parts and later on paste them all together. However, it seems difficult to join them all cleanly.
Good luck!