It’s just one option from my head - to use TeamCity and have different build scripts that know how you want to break the tests and install them as separate projects (each section), and then configure n agents, which can be done using Amazon EC2, which allows large values of n.
Since you probably want more than three agents, you do not fall into your free territory.
Some assumptions: there is no common database that links all these tests together, and all tests can work independently of each other.
If there is a common database, this makes things much more complicated, since you will need a database for each agent so that the tests do not attack each other (and, of course, computing power for EC2 would probably be impractical).
Yishai
source share