Oh man—all kinds of stuff. The visible bits are an iceberg tip. And it all runs in a Racket process on a single core.

That mammoth thread at least tells us where the brick wall is: at about two 'who is hiring's.

How much does calculating the approx time of posting for every comment add to the total load?

If it’s significant then you could ship the UTC posting time to the client & calculate the displayed value in Javascript & eliminate that particular part of the load at least, which in turn might enable more caching internally.

(You’ve probably looked at this already though.)

The last time I profiled the code, the timestamps weren't expensive.

Good that nobody ever tried to combine the two, i.e. who is hiring and who wants to be hired, earlier.

Is it the total number that crushes things or the high level of nesting?

The total number. We actually stop rendering comments past a certain amount of nesting; ever see a 'more' link where 'reply' would normally be? That's what that is.