I have selected this chip for use in a project of mine.
It has some very nice features and it is to replace an older HP
(now Agilent) part that has been available for many years.
Because the inputs have shift register filtering I don't think you have any kind of noise problem. The count logic internally specifies seven clock cycles maximum propagation time; and further states matter-of-factly, that "Illegal state transitions caused by faulty encoders or noise severe enough to pass through the filter will produce and erroneous count". In testing this part in my application I have not seen any bad behavior, but I will say that I have seen this in even high-end servo controls when they do repeated rapid reversals (such as in peck-drilling). The quadrature
detector uses edges of the A-B phases so when resonance or rapid reversal causes the encoder to roll back and forth across an edge in 1X decoding this causes what they refer to as an "Illegal state
transition". In 2X decoding both edges of one phase are used and in 4X decoding both edges of both phases are used. Ironically if static decode logic were used in 1X decode you might not see the problem!!!