a) Without having access to the numbers, (Skummet) startup is ~5 times as slow as the Java version. That is a noticeable difference. Plus, even Java apps sometimes tend to take a second or more to start up if they're reasonably complex.
b) The "Lean Clojure works" statement is derived from comparing _only_ the startup times. The fact that the task time is around two times Java's for Skummet is a big thing. Ignoring startup time, and without further data/tests, one could argue that Clojure Android apps run twice as slow as Java ones... that's a big no-go given the high amount of sparkly UI features of most modern apps. Or at least a no-go for having a Clojure-only stack?
(Skummet's probably in its early stages and I'd expect improvements but ... from those graph I'd argue 'Lean Clojure' does not work (yet?))