I'm sorry, did you claim poor tuning?
look at this graph: https://alvinalexander.com/sites/default/files/inline-images...
from: https://alvinalexander.com/scala/graalvm-native-executables-...
Graal is very clever, but that memory example is nonsense. I just tried the Find example they give and whether the graal version beats the java version is completely dependent on Xmx.
Obviously I don't have their dataset, so it's not exactly the same test.
edit: In fact the numbers in the blog are so far off mine they may accidently have been testing /usr/bin/find.
Startup times are not an issue on servers as processes are typically long-lived anyway. It is a bit of an issue on serverless (lambda etc.) but that’s a relatively new thing. High memory usage is generally poor tuning - if you give the JVM 1gb it will use it and avoid GC until it needs to. Sometimes it’s poor coding too, I’ve seen web APIs accepting files as base64 encoded json strings which is horrific server side.