CPU performance-wise, how much faster the execution time in Google App Engine's backend compared to the one in non-backend?

We are developing a prototype app on top of Google App Engine's platform, but some part of the code requires quite intensive CPU processing and exhibit abysmal performance. We tried to optimize the code, but we are also trying to explore the throw more hardware approach for our optimization purposes. :-)

The official documentation from Google seems to emphasize the features: backend allows code to execute more than 30 seconds and you can have always-on instance using backend, but does not give clear outline on how much faster we could expect our code to execute compared to using a regular, non-backend, environment. Looking around the web does not help much either.

I understand that using the higher classes of backen开发者_StackOverflow社区ds (e.g. class B8, 4.8GHz virtual CPU) would most likely give me much faster execution time, but I'm curious (and hopeful!) whether even using the lowest class of backend could give us much faster execution time.

So, I'm hoping the crowdwisdom in stackoverflow could share their experience regarding how much faster Google App Engine's backend compared to the non-backend environment, performance-wise?

Now that we are several months in the future, Google had released App Engine 1.6.1 and adds the capability to adjust frontend instances to use different classes, things has become much more transparent.

It now could be concluded that the default front end instance is equivalent to a virtual machine having CPU capacity of 600MHz. This is equivalent to the CPU capacity of the lowest-class backend instance.

So, even the lowest-class backend instance would not give me a performance boost compared to the default frontend instance class. The good thing is that we now could choose a higher-class frontend instance rather than relying on higher-class backend to execute CPU-intensive code.