What is the best freely available face-detector implementation?

OpenCV has a nice implementation of Viola-Jones style face detection. (Basiscally this uses a boosted cascade of Haar features -- allowing for fairly robust and computationally efficient object detection.)

http://opencv.willowgarage.com/documentation/cpp/object_detection.html

The code is open-source, although there may be IP issues to consider if you want to use this algorithm in a commercial app.

Yann LeCun has some supervised deep convolutional neural nets that seem to work really well on the face detection task. For instance the LeNet5 architecture seems to work fine for this. AFAIK there is ready to run push button implementation of such a model but it should be possible to train one with either theano or eblearn and a faces databases such as one of those referenced here or there.

There is also a LGPL Java implementation of the Viola-Jones algorithm available. It further comes with the ability to load the cascade model files from the OpenCV project.

If you want to think a little outside the box of Viola and Jones(!) you should definitely look into t-SNE methods (and neighbor embedding in general) Although I kind of agree that viola and Jones is hard to beat. There are MATLAB and Python implementations available on that website

Try libface (http://sourceforge.net/projects/libface/). Also I've modified one of their examples to save the coordinates of faces detected from the input images (http://github.com/andrewharvey/perlmisc/tree/master/detectfaces/).

This is not freely available, but you should definitely check out http://pittpatt.com/
Their online demo of their face detector is pretty impressive (I love the scroll bar which lets you make precision-recall tradeoffs)
( Maybe their product comes up with a reasonable price-tag or they have big discounts for academia )