Yikes! 20% of quota in 1 second! Holy crap.
I needed to do something similar, where I need to basically run an A-
Star search algorithm to solve an optimisation problem on GAE. Now
I'm generally expecting the datasets to be fairly small and therefore
the runtime of the search should be small, however, I'm now very
worried about scalability if a single second of runtime is 20% of your
quota.
Sounds like it might get very very expensive over time. And breaking
up the computation into chunks-per-request shouldn't make a difference
to the total CPU bill ...