Yes there is light at the end of the tunnel and it isn't the train. The final runtime pieces of tiered compilation got putback into mustang. We don't build the tiered system though. At the moment it is buildable only on solaris but that is only a makefile issue all the runtime pieces and compiler pieces are present. That isn't to say that it doesn't need lots of tuning. There are lots of policy decisions that are being worked on but those code changes will be trivial compared to what has been done during mustang. Unfortunately I was quite fast enough. The intention with the recent putback is that I (we) expect that there will be pressure to backport it into an update release of mustang so the last major code changes got put into place now.
I have done some benchmarking of the tiered system on x86 and sparc (no amd64 since we need a client compiler). At the moment the numbers are encouraging. Typically we see startup performance as good as with client but final performance is either a little lower or a little higher than using -server. I'll be part of a group presentin at Java One on "New Compiler Optimizations in the Java HotSpot™ Virtual Machine" and I hope to have some more performance data there but we'll see. It will certainly cover some of the work in getting tiered ready,