Now, my question would be whether it is common practice for manufacturers to include a buffer on headphone amps before the gain section.  Is it always done, or is it sometimes done and sometimes not?  Could this be why some amps receive wildly different reviews, or is "synergy" and "taste" still the most likely culprit?

This is an interesting question, but requires a little thought about the nature of impedence and AC signals.
Unfortunately, it's not quite as simple as just adding a buffer stage that provides the right impedence on both sides.  Impedence is not really a single number, since it is an AC phenomenon -- most of the time it varies with frequency, especially in output stages.  Transistors and Tubes differ in their linearity and ability to deliver a flat power curve into varying impedence loads.  Different input and output topologies will affect this power transfer to a greater or lesser degree.  Buffers are active devices that cannot be counted on to be entirely linear across a wide range of AC frequencies, plus as you point out, many of them tend to have a sonic imprint they add to the music.
For an alternative, think about transformers.  Any transformer is an impedence matching device which can be used to improve the power transfer across audio frequencies.  Different turns ratios or taps on the transformer can be used to tailor the impedence match to optimize the power transfer.  But transformers that are really flat across the entire audio spectrum are expensive. 
On the input side, some designs couple the headphone amplifier to the source across a capacitor, while others apply the signal though a transformer, or even use a buffer circuit as you suggest.  Each connection approach has tradeoffs.  Usually the designer will tailor the tradeoffs to maximize sound quality (or minimize parts count) -- based on price point and priorities for production.  I personally am partial to dual-differential fully balanced input stages, using transformers to couple.  But I've built amps that used other approaches with excellent SQ.  Anyway, it is not as simple as slapping in a buffer stage to solve all the ills of impedence matching.  Wish it was...
On the output side, transformers can work great as impedence matching mechanisms to link the output devices to the load.  But again, they all vary somewhat in their performance at different frequencies (altho their limits may be well outside audible frequencies).  Or in another case, an OTL (output transformer-Less) design directly couples the power devices to the load.  If the impedence match is good, and the design carefully done, the amplifier will be operating its output devices in the most linear part of their operating range, where every increase or decrease in inputs yields an exactly proportionate output.  But the load can complicate that as its impedence varies with frequency.  So nearly every OTL amp has a fairly narrow range of load impedences where it will sound its best.  Fiddling with the input impedence doesn't solve problems of varying power delivery across different output frequencies.
The good news for Audez'e users is that the LCD-2 is mostly resistive in load, barring cable capacitance, etc.  So these cans provide a fairly nice load to most amps that can handle a low impedence, provded they can deliver enough current.
Designing this stuff so that it sounds outstanding, rather than merely good, is an art form.  It is not just all-so-simple to make sure all of the possible input frequencies are along the linear portion of the transistor/tube, given ALL of the possible output voltages from the previous stage.  At the same time, the designer is considering the bias voltages being used on each device, feedback, and then coupling to next stage, or to a load.  Remember that the source can output mere millivolts all the way up to 3 volts, or even 5 volts in some instances. The amp needs to be linear across this range of voltage, at every audio frequency.  I am in awe of the great audio designers, like Nelson Pass, John Curl, or Alan Kimmel.  It is quite an accomplishment when it all comes out sounding spectacular:  selecting the right devices and setting up a circuit topology that operates them to best effect, while not overdriving them (for longevity), and not under-driving them (for noise elimination & non-linearities) and avoiding the use of band-aids like negative feedback loops.  I've spent a lot of time looking at circuit diagrams and rebuilding vintage tube amps from the 40's, 50's and 60's.  It is simply amazing all the different solutions and circuit topologies that have been used simply for audio amplification.  While there are many "standard" topologies that are heavily used, there are often really creative solutions that break the mould and accomplish fantastic sonic results (see for instance the way David Berning couples audio signals by piggybacking them on top of RF signals!).


Originally posted by Frank (DeadEars)