That gets things down to 0.183s, which is certainly an improvement. But Python manages 0.033s and Guile manages 0.022s, neither of which need a precompile step. I feel like Racket needs to spend some time optimising this.
My understanding is that Racket has been doing a lot of work on overhauling its interpreter and internals recently, so I'm hopeful this will improve somehow, but it's unclear to me if they're aware of this issue or consider it a problem.
The compiler was rewritten and now Chez Scheme is used as part of the pipeline from Racket to assembler. With respect to improving the startup time, the low hanging fruit has already been picked.
If I understand correctly part of the problem is the number of modules that needs to be loaded from disk at startup. On my computer the first run is slower than subsequent runs due to caching. Maybe an "compile several modules into a single file" approach might improve things - but it's not trivial project.
My understanding is that Racket has been doing a lot of work on overhauling its interpreter and internals recently, so I'm hopeful this will improve somehow, but it's unclear to me if they're aware of this issue or consider it a problem.