The managed code in your organization must have at least 75% coverage to have been installed originally. When you use "run all tests", without attempting to deploy your own code, what is the result? Low code coverage from an installed package typically occurs because of the test aborting. This can happen when the administrator adds some code or configuration changes that affects the testing process.

Consider this scenario:

public class ManagedTestClass {
  public static void testMethod test() {
    Account a = new Account(Name='Test');
    insert a; // Trigger a managed Trigger.
    // do some more stuff with account a
  }
}

The developer creates a new account named Test, and when the module is installed, there is 99% coverage. Now, the system administrator decides that no account should ever have the name "Test" in their name, so they add the following validation rule:

 CONTAINS( Name , "Test" ) 

You can imagine what happens next. The validation will fail on the "insert a" line of code, thus wiping out the rest of the test and all future tests. If that happened approximately half way through the test process, the coverage of that managed code would hit 1% (because it literally failed on the second line of the test), and when averaged with the rest of the code, might easily drag the 99% coverage to 40-50%.

In this example, the system administrator for the organization the code was installed in was not aware that he "broke" anything when they made the new validation rule. In fact, the system could still be running 100% correctly, and only the test would fail.

Check your debug logs or deploy logs (i.e. Eclipse debug logs) to determine what broke, and then you can work on fixing it. Either your code is interacting with the managed code, or a setting in your organization has broken part of the managed testing.