Actually, CPU usage doesn't translate linearly from Axe III to FM9. TBH, 56% CPU usage in the Axe III is about 72% (just spitballing the estimate) on the FM9. The question you've raised is actually a very good one, which I'll ask on the FAS forum right now.
CPU usage is dependent on what type effects blocks one chooses. For example, a standard delay uses less CPU than their PlexDelay or MultiTap Delay. Huge washes of reverb also use more CPU, but not as significant as delay usually does.
EDIT: The answer to your question isn't something you can provide a blanket response. Because CPU usage in amp, cab and effect blocks vary greatly dependent on the blocks themselves, and how many are implemented, the increase from Axe III to FM9 can vary anywhere from zero to 25%. Sorry if this doesn't provide a definitive answer. The clue may be to simply try experimenting with either FAS device and see what happens. You'll discover quickly enough what each device will tolerate or not.