A JRE has to uncompress each class to interpret or JIT it. Each JRE instance
has it's own copy of the uncompressed bytecode.

If the jarfiles used no compression, the JRE could directly mmap() the
bytecode in MAP_SHARED/PROT_READ mode; this would allow it to be
directly shared by other JRE instances without any other IPC.
It also avoids using memory for both the filesystem image (compressed)
and a heap buffer containing the uncompressed bytecode.

If disk-space is at a premium, a compressed jarfile could be used, but
if the jarfile contains no compressed entries it could be simply
mmap()ed and shared, thus reducing memory footprint.

Clearly it would be even better if JIT-ed code could also be shared
or even persisted, but sharing byte-code would be a great start.