That's very sad, didn't the typed racket guys say the exact opposite at one point?
Efficient gradual typing is still an active area of research. Also, see section 5.1 on the validity of their findings.
They claimed 100x+ speedups by using typed racket vs untyped.
And they probably did get these speedups when they replaced a large body of untyped code with typed code without relying on untyped libraries. The issue is that gradual typing has to work, well, gradually. Gradual typing won't be viable as a speedup strategy until this is improved. If runtime costs are too high, gradual typing won't be viable as a correctness strategy either.