Actually APM as it is defined by legacy trace & diagnostics solutions is ill-suited to this particular use case (dynamic resource provisioning) in the cloud. The only time this looks good is when you're half asleep in a meeting room during a product presentation were the dumbest of web apps is being scaled by such simplistic tooling. The best APM solutions can do today is collect information presented to user. Acting on such information derived from the data collected is a completely different ball game if the goal is to deliver actually performance improvements in an cost and resource management efficient manner outside of just flogging vapor ware.

That said I think there are possibilities in streamlining some aspects of capacity management once we drop these legacy trace/log/diagnostics tooling in favor of a solution that is aligned much more closely to the nature of the problem - resource metering. This is an area I have been working on for a number of years even prior to the cloud washing scam sweeping over various products in the Java space.

If you want to glimpse at the future then you should check out a number of blog entries I have published with less talk and more substance hopefully.

http://williamlouth.wordpress.com/category/cloud-computing/

My vision formulated a while back can be found here:

http://www.jinspired.com/products/jxinsight/meteringthecloud.html