I've been playing with this, and I think it's the right way to go. In particular, we'll keep using trial for our tests (although meejah tells me that py.test knows how to handle Deferred-returning tests, and has some nice composable "fixtures", so maybe we should migrate towards that). But we'll use tox as the frontend, and our docs can say:
To run the unit test suite from your local source tree, run:
tox
That will create a virtualenv (one for each supported version of python), install your local source (along with all dependencies) into the virtualenv, then run the tests. That means you don't need to create your own virtualenv for testing purposes (although you will if you want to actually run tahoe). The docs can provide more specialized instructions for running specific tests, installing development versions of dependencies (e.g. test current Tahoe trunk against current Twisted trunk), and getting code-coverage data.
Something that I care about a lot: a very fast test-edit-repeat cycle is still possible, by populating the virtualenv with --editable, then running something like .tox/py27/bin/trial allmydata.test.test_foo.
