sensor size - Why do depth of field calculators show *more* DOF for larger formats with the same lens parameters?

I thought I had a pretty good seat of the pants understand of DOF. But I've been playing with DOF calculators and it's giving me results that don't line up with my practical experience, so I'm wondering where I'm going wrong here. This is about the relationship between film/sensor size and DOF.

Let's use some fixed parameters (keep film size variable) for the sake of argument:

focal length: 100mm, 
aperture: f/8, 
subject distance: 20 feet.

I know (from experience) that 35mm film has much more DOF than 4x5 film with those parameters. But the DOF field calculators I've tried are telling me the opposite, that the larger film/sensor will have greater DOF. The DOF calculators I'm looking at are http://www.dofmaster.com/dofjs.html and http://www.cambridgeincolour.com/tutorials/dof-calculator.htm, and both are telling me that larger sensors/film will have more DOF all other things being equal.

The Cambridge in Colour DOF calculator says DOF (with above parameters) for 35mm would be 6.29 feet (makes sense to me) and for 5"x4" film 34.6 feet (doesn't make sense to me).

So I'm missing something here in either my understanding of depth of field or the terminology or my understanding of DOF calculators. Can anyone straighten me out? What am I missing?

Answer

Here's what you're missing: that larger formats have less depth of field for the same framing, not at the same focal length. A 100mm lens is much wider on medium format than it is on 35mm film. If you keep that and the aperture constant, DoF will be identical assuming you print with the same enlargement (that is, the medium format print will be much larger). If you instead enlarge the 35mm print to the same size, you also enlarge the blur, yielding the decreased DoF the calculator is showing you. (It has these assumptions hard-coded into its chosen values for circle of confusion.)

Instead, adjust the focal length so that you have the same framing. If you hold that constant, you will find that your original understanding still holds: greater depth of field for the smaller format (even with greater enlargement). More background on this at previous question Can a smaller sensor's "crop factor" be used to calculate the exact increase in depth of field?