There is also an impact on memory use and startup latency imposed by JITs vs AOT. None of the JS VMs so far seem to do any kind of snapshotting. Since the types

There is also an impact on memory use and startup latency imposed by JITs vs AOT. None of the JS VMs so far seem to do any kind of snapshotting. Since the types collected by runtime profiling are likely to be the same for most runs of a well-behaved app, it seems like you could send down the profile to the client JIT to allow it to converge quicker.