How much do we care about evaluation performance? What are the bottlenecks?

[effectfully]

I've been told before that performance of the evaluator wasn't considered a priority, so after some initial work I personally largely stopped pursuing optimization and have never touched these ideas for example (plus the performance of builtins can still be improved further). This was some time ago though and I haven't asked about that in a while.

However after I asked the community for feedback, it kinda became clear that performance should really be a priority. Hence I'm creating this issue where everybody is welcome to express their views on how important performance is and which parts should be targeted first.