CPU has not been the bottleneck in steemd - but with the release of appbase it can be. Making CPU the bottleneck is actually a good thing because then it becomes horizontally scalable based on usage. It gives you a metric that you can actually scale instances on - previously it was based on the number of requests, which is difficult to guess at because certain requests can take longer than others. To answer the question directly, appbase steemd can handle an extremely large number of requests before becoming bottlenecked by CPU.