Stepper vs. Servo control

I've been following the latest discussion about low-cost servo amplifiers,
which got me thinking about step/direction servo systems such as Dan Mauch's
(the Servolite?):

If all moves in a CNC system are in the end very small linear point-to-point
moves, due to interpolation, contouring, etc., then what is the difference
between a system that uses step and direction output to a step/direction
servo driver such as the one Dan Mauch sells (Servolite?), vs. the more
common motion controller/servo amp setup. Assume the resolutions, motors,
and encoders are the same, electrical noise isn't a problem, and we're
talking about a tacho-less system. It seems to me the main difference would
be in the PID filter, and probably something to do with "feedforward".
Whether you keep track of "commanded" position in the PC, on the motion
controller board, or out at the step/direction driver doesn't seem to me
like it should make much difference.

I'm asking because software like IndexerLPT can generate over 100,000
steps/sec on a slow Pentium, doing all of the interpolation for several
axes, with unlimited lookahead. If you wanted machine resolution to, say,
.00025", and designed with 10X that for the encoder/electronics resolution
(.000025"), that's 40,000 steps/inch. That would allow you to run at up to
2.5 inches/second (150 ipm) which sounds like plenty for what most of us
want to do. Cut the electronic resolution down, and the bottleneck would be
somewhere else. The lookahead, in my view, is what would save this system.
Being able to anticipate and plan out all accelerations and decelerations in
advance would, I hope, remove some of the burden on the PID control system.

It seems very attractive to me to be able to eliminate the need for motion
control boards (ServoToGo and the like), and talk directly to step/direction
servo drivers. Status and E-stop concerns could be easily reported to the
PC with a few more bits. The PC would have to expect that the motors were
where they were commanded to be (or on their way there) unless the status
flags indicated a problem.

Any comments on this approach? Any reviews on current step/direction servo
drivers and their limitations? Am I just crazy too?

Carlos Guillermo