I am looking at this as well.  Any progress so far?

I think the possibility of missing pulses will always be there depending on the number and speed of the multiple encoders attempting to be supported.

My testing agrees with you.  The board design is finalized and I am finishing up the documentation.

First, I am not sure how magnetic is better or higher tech than optical.  Second, my current design would work fine with any encoder output including the ones that that manufacturer sells.

I am using a similar chip, but not the iC Haus one.  I have been bitten by their some-what-quick obsolesce cycle in the past.  Their chips *are* nice though.

If you do not care about direction, a pulse wheel encoder (rather than quadrature) would be a less expensive option.  Also a simple hall sensor setup a'la bicycle speedometer or gear tooth sensor might work for you.

Without knowing your specific application it is hard to say.  As mentioned above, your examples would not work for continuous rotation.  The following encoders have ease of integration as they 'snap' on to existing shafts: http://www.cui.com/Catalog/Components/Encoders/Incremental_Encoders/Modular

It could be that was it.  I attempted every variation of software I could find, but everything lacked repeatabillity.

I hear you, but empirical testing with the interrupts show lost pulses due to collisions.

 I see how I was not clear.  I am hardly 'detecting rotation' nor just reading an encoder in 2ms, but doing joint angles on 6DOF articulated arm and forward kinematics to ultimate position in that time.  However, if you can get repeatable results with zero dropped pulses (on a 2048-50000 ppr encoder) using any of the 5-6 methods floating around the Arduino forums please let me know.  I had pulses dropping out on every method I could find with only one encoder moving at about 256 pulses a second.

I am doing a project that utilizes multiple rotary encoders.  I had repeatibility problems with every solution I found in my online searches that read quadrature directly into the Arduino, even with just one encoder.  This was probably due to the rate of the pulses in my particular application. 

So I designed an encoder shield that can handle up to 8 encoders simultaneously and the Arduino seems to be able to read and report the counts via serial to the PC in about 2 ms.  Anyone else interested in such a thing?