My experience is different. I am using more than 1600 records and and there are four tables are connected. But I am getting reasonably good performance as of this time.

Performance depends upon many factors so it is difficult to conclude just based
on number of records.

However, it is true that Memento does not have good performance and at some point of time it may give performance issues.

Agree, this is the only thing holding me back from deploying at scale in my organisation. I use it for small point solutions but it doesn't scale well enough for general use.

I found a bug in this concern,
I have several JavaScript fields in my library which are used as Name, Description, Status and Colour of the library.
In All these fields, I have turned off realtime execution, but still loading the library takes 10-30 seconds for 100+ records, but as soon as I set these js fields as Regular Fields, library loads instantly!
This means that inspite of realtime execution off, memento is executing the JavaScript in js fields if they are set as name/description etc.
This behaviour should be fixed and it should use the static value as result of last execution.
We can create an action script in library view to recalc all lib().entries() in library if we want to manually refresh all js fields.

This would improve the load time of libraries drastically.

Definately needed, maybe defer calculations etc of older entries on opening to make Library open faster

Definitely still need this!! (2018)