If a gem has rails dependencies, do you think it is better to write the gem tests in a way they can be run standalone or run them under a rails project?
I'd say it depends on the kind of dependencies the gem needs. Eg. if it's just the ActiveRecord it's quite easy to include it into your test suite. In more complex cases you can always mock some of the needed functionality. In a really complex cases, creating a test app is better than nothing (IMO).
A gem should be a piece of code which acts stand-alone. Otherwise it is part of the application, so the tests should be created stand-alone as well. In this way others (hypothetically) can perform the tests as well. If the tests depend on your application others cannot test your gem.
Furthermore when you want to test your gem it should not fail because your application is failing. After your gem passed the test, you can test the application knowing that your gem is functioning well (assuming that you tested everything).
Whether the gem is depending on Rails or not is not an issue, since Rails also has been tested (and you can assume it is working properly). So these dependencies do not (/should) influence your gem.