I agree, but the paper you link has more than one year. Now the benchmarks are much better, and in most cases Racket CS is faster than Racket BC. I think the even point was a few month ago, but there were too many bugs (small bugs, in weird corner cases, but important enough to not switch until they were fixed).

If you don't mind crappy graphics with numbers pulled out of the air https://imgur.com/a/9r0aJzd The idea is trading a slower development speed for some time, to get more speed and features in the long run.

Perhaps the change in the slope in BC is too much, and the current speed of CS is greater than the imaginary speed of a version of BC in an universe where BC didn't happen. I hope the crappy graphic in Paintbrush is enough to make it clear that the numbers are pulled out of the air. Anyway, one of the hard fact of the blog post submitted here is that in most cases the current version of CS is faster than the current version of BC.