slow-checker or slow-parser.rkt easily fails due to out of memory when running in parallel (j > 10)

[capfredf]

I've found that either of these two tests fail 9 out of 10 times when I run the integrated tests with all the threads via '-j n` (n = 12 when I was on my mac or n =16 on my Linux laptop)

running the tests indeed uses more memory when n is larger, which is expected: n = 1, the peak memory usage: ~800M n = 16, the peak memory usage: 5G

It is more like a place + sandbox issue rather than a TR issue, but I just wanted to first posted here and see if I reproduce the problem with much simpler code.