Richard Aiken wrote (of neuromuscular blocking drugs):
 > If you're delivering the toxin to a limited set of specific neural
 > intersections - because you're using tiny self-directing robots rather
 > than random distribution - shouldn't that reduce the required dose
 > rather significantly?

No; the distribution isn't random.
While there are so many molecules that statistics and probability can
come in to play, drugs move between compartments due to concentration
gradients, solubility coefficients, binding to carrier proteins, and
ultimately their affinity for the target once they are in range of it.

So multiple physical phenomena are involved: bulk flow (e.g. blood
circulating through target tissue), diffusion, solubility and the
reaction kinetics of binding to carriers or targets.

The 'limited set' of neuromuscular junctions is large (~10^8) as each
skeletal muscle fibre has a accompanying nerve ending.

The required effect site concentration to produce 50% blockade (EC50) is
on the order of 100-1000 nanograms/mL, or 10^(-6 to -7)g/mL.

The functional volume of the effect site is small - which gets you back
to the same order of magnitude [~10^(-3)g] posted previously.

You raise another problem with nanotech: how do the tiny robots
'self-direct'?

Chemical or temperature gradients?
Recognition of (cell and tissue) surface molecules?

They are too small (micrometre and smaller scale) to have
electromagnetic or sound-based sensors.

Rob O'Connor