While it is easy to imagine selective-detection effects that mess up the results enough to invalidate the test at the level of the inequality, it is very, very difficult to maintain all the physics required for precise, detailed agreement between theory and experiment of the kind that Aspect and others have shown. Here is an example of a "local realistic model" that reproduces the quantum mechanical results for in-time coincidences, but completely messes up any number of auxiliary measurements: http://www.tjradcliffe.com/?p=590
So while I'd love to see a modern version of Aspect's work using state-of-the-art entangled photon sources and the like, the likely reason it hasn't been done is that the odds of it revealing anything new and different are trivially small (but not zero, of course!)
"Chaotic Ball" model, local realism and the Bell test loopholes
http://arxiv.org/abs/quant-ph/0210150
...any thoughts?