Ugly special casing requires code that would detect those special cases. This adds overhead in terms of processing cycles and code. Most arrays, majority of the time, store a tiny amount of information. Less than a dozen entries. Doing that sort of detection is massive overkill and would actually hurt performance. You want to speed things up for the real world. Not artificial benchmarks. How many times in your code are you really going to need 100,000 integers in memory? Probably once in your entire lifetime, if that.

I use much larger integer arrays all the time but that's beside the point. The question is whether or not it makes sense to make dynamic language runtimes as efficient as possible. I think it does, not because it necessarily helps all or even most existing programs, but because it enables people to create a much wider range of software with a language they already know. Many modern websites wouldn't be possible with JavaScript engines from 1999.

And if you optimize at all, memory usage must be a top priority. Lack of memory is a lot more costly in terms of performance than a couple of C based heuristic checks. Creating and then garbage collecting an array full of individual integer objects doesn't just use a lot more memory, it's also orders of magnitude slower. The array doesn't have to be very large for that to matter as there may be large numbers of arrays.