You misunderstand me...it has to read and decompress (jars are basically zip files) those 50MB are so (just for the system jars) on every startup before it can even think about validating the cache
Sure, however you are still speaking of a specific implementation.
And which implementation that you seem to imply exists DOESN'T suffer from this seemingly fundamental issue? You built this strawman, now defend it or stop waving your hands.
Just read my OP carefully and you will find you answer.
No, read MINE carefully. Even loading cached data isn't free.
Of course not, all the code and data that go into the processor needs to go through the processor cache first.
My OP comment has an example of a JVM that doesn't read jar files as you mentioned, while executing native code stored in a JIT cache present in the hard disk, between invocations.
Google is your friend if you look for the respective documentation. Or, who knows, you could even download it and play around with it.
I'm out, please turn the lights off on the way out.
JVMs that cache JITed code don't spend time loading the classpath if the cache is still valid.