As noted, GD&T can help make your tolerancing fit the function, but like any versatile tool its effectiveness depends heavily on the user. Proper choice of critical dimensions is essential to reducing tolerance stack-up. Manufacturing must also recognize the dimensions as critical, and adjust the process and tooling to suit. I highly recommend that ANYONE working with mechanical designs be trained in GD&T, but be forewarned: tolerance stacks that involve GD&T controls are a bit trickier to calculate.
Statistical tolerance stack calculations can be used to allow larger individual dimension tolerances, but only if the core assumptions are understood and handled. The method assumes a normal distribution, centered on the nominal value. A certain defect rate is expected, and can be estimated. Basically, if you don't have suitable SPC data for your parts, or if you have less than three stacking dimensions in the calculation, just use worst-case tolerancing (or risk suffering unacceptable defect rates).
The best resource I've found on these issues is the Dimensioning and Tolerancing Handbook (Drake, 1999). It includes chapters on GD&T, one-dimensional tolerance stacks, multi-dimensional tolerance stacks, and Six Sigma tolerance design. Amazon.com carries it, but if you get a copy, treat it gingerly - the binding on this edition is horrible, and mine is already falling apart.