Fixed point is very handy for accumulating small bits to gain resolution.
An example using 16 bits, where the top 8 bits (H) are the integer component and the bottom 8 bits (L) are the fractional component.

Code:

HHHH HHHH . LLLL LLLL

It's quite easy if you define a new type using a union/struct...

Code:

typedef  union
{
  struct
  {
      byte low;
      byte hi;
  };  unsigned  all;
} fixedPoint;

fixedPoint Accumulator;

Adding to Accumulator.all rolls from low to hi.
Access to the fraction component is by Accumulator.low
Access to the integer component is by Accumulator.hi

The SAM3A4C and SAM3X4C are the smaller (100pin/256k) cousins of the 144 pin SAM3X8E on the Due.
Being in the same family would make quick and dirty ...

Rob is correct, that is the case with ARM....

Some steppers are 1.8 degree, 200 steps. Some are not...
The Epson steppers might be 3.6 degree, 100 steps.

You will need two H bridges to control a bipolar stepper, one for each coil.
What driver are you using, where is the schematic?

Is this it: http://arduino.cc/en/Reference/StepperBipolarCircuit
In this case the L293D is a dual H-bridge.

If it just vibrates then you have some motor wires mixed up...
Or you might be trying to move it too fast with too little current.

Try reversing one coil and slow it down...

Sounds like the load cell is only half a Wheatstone bridge, a resistive divider, you can verify this with a multimeter..
It would be wise to add a divider for reference, that way small fluctuations in supply will be canceled...

IIRC, all the legs were in series.
For hookup, you will want to research Wheatstone bridge, a couple of links:

http://www.maxim-ic.com/app-notes/index.mvp/id/3426
http://en.wikipedia.org/wiki/Wheatstone_bridge

The sensors should have four wires, with two active elements and two passive.
Essentially it is two resistive dividers where one is reference and the other changes.

The output of these two dividers is treated as a differential signal going into the AD623...
The instrumentation amp gives an output that is the difference between the two inputs...

An instrumentation amp is a good start, the AD623 comes to mind, unlikely to find that at radio shack...
Also there are complete chips that include AC excitation for the load cell bridge if you want to get fancy...

Here's an example from 2008:
http://krazatchu.ca/2008/12/31/data-logging-weigh-scale/

Yes, the load during movement...
If the load/force is constant and the motor power is constant then the speed will be relatively constant given that the heat is negligible...

If the load is dynamic and you want to maintain constant velocity then an encoder is highly recommended...

Did someone say assembly?
http://n0m1.com/2011/12/21/inline-assembly-a-fast-quadrature-decoder/

Yes, use a low side current sense resisitor, amplify with an op amp and read with the ADC...

That's a big servo motor, 9.3 amps at 48 volts and 400 watts...

Not an easy task to run PID, an encoder read, as well as brushless commutation in real time on an Arduino...
As well, the PCB layout and motor driver will be critical at these power levels...

I've dabbled in brush motor servo drives which are tricky enough, I wouldn't even attempt brushless PID on an Arduino...
Starting from scratch without experience, I suspect your looking at least 6 months of learning and work...
I'd recommend an off the shelf driver for them, this one may work and is reasonably priced:
http://granitedevices.fi/index.php?q=servo-drive-vsd-e

No experience with this specific chip but the interface is CNC standard opto-isolated step and direction.
One pin for direction and one pin for step, send a pulse on step and it moves in the direction defined by the polarity of the direction pin.
It has microstepping as well, which will require that you multiply by this number of steps...

You can operate the 48v motors on 36v if the current is available, they will just run slower and cooler...
The Adafruit motor shield is for brush motors not brushless...

I strongly advise you to consider brushed which are a lot simpler to drive.
Driving a brushless requires commutation which in turn requires knowledge of the motor position, either through back EMF or Hall sensors...

As well, what level of control are you aiming for?
Position, constant velocity, constant torque?

You can generate much faster frequencies with pwm but you end up losing resolution by decreasing the bits/top value...

Normally the atomizer transducers use a self resonant circuit to maximize power transfer.
A Google image search for "ultrasonic atomizing circuit" gives lots of examples like this one:
http://www.seekic.com/circuit_diagram/Basic_Circuit/Medical_ultrasonic_atomizer_1.html

A few years back I experimented with a 200w ultrasonic traducer, I used phase angle to track the resonant frequency:
http://krazatchu.ca/2008/04/17/ultrasonic-resonant-tracker/