I have not gone through all of the algebra of the Hinton-Rosenbluth
Sherwood poster and earlier memo, but here are a few of my own comments on
why I think Dimit's interpretation is correct.

Physically, I believe what is going on is that it takes a bounce-time to
set up the collisionless "neoclassical" enhancement of the polarization
shielding.  I.e., in the gyrokinetic equations, the usual ion polarization
shielding comes from the (1-Gamma0(kperp*rho)) term in the
gyrokinetic-Poisson Eq, which for small kperp*rho gets expanded to involve
kperp**2*rho**2.  On a long time scale, one also can average over banana
orbits to get a similar shielding effect from the distortion of the banana
orbits by the radial electric field.  This enhances the shielding by a
factor of 

D_neo/D_classical = (n_t/n) * rho_banana**2 / rho**2

where n_t/n is the trapped particle fraction and accounts for the fact that
it is primarily the trapped particles which have such large orbit
excursions off a flux surface.  Here D is the perpendicular plasma
dielectric (NOT the particle diffusivity), and the total D = D_neo +
D_classical has components due to the classical and neoclassical ion
polarization densities.

Hinton and Rosenbluth's Sherwood poster focuses on times long compared to
the bounce-time, and so it bounce-averages (see the comment before Eq. 30,
or the comment at the beginning of the hand-written appendix that omega <<
omega_bounce is assumed).  Thus it can't say anything directly about
dynamics which occur on time scales comparable to or shorter than the
bounce-time.  Thus the statement in the Discussion section (Sec. 11) that
"it is clear from our solution that transit time damping of poloidal
rotation does not occur" seems to me to be an overstatement.  It depends
somewhat on how the system is being driven (or what the initial conditions
are), but is seems clear to me that the bounce-time required to set up
neoclassical polarization shielding looks as if it were a transient
"transit-time" damping followed by an undamped residual component.

However, we can use Eqs. 56 and 57 of Hinton and Rosenbluth's Sherwood
poster to calculate the ratio of the neoclassical to classical radial
polarization currents:

j_neo/j_classical = 1.6*q**2/eps**0.5

(assuming that grad(psi) is defined as B_poloidal*R without any factors of
2*pi). Except for the factor of 1.6, this agrees very well with the above
formula for D_neo/D_classical, taking rho_banana=rho*q/eps, and
n_t/n=sqrt(eps).  If one set up an initial phi due only to the classical
polarization density, then after several bounce times the final phi should
be smaller by a factor of

D_classical/(D_neo+D_classical) = 0.6*eps**0.5/q**2 / (1 + 0.6*eps**0.6/q**2)
				= 1-K_p

where K_p is the quantity in Eq. 9 of the earlier memo by Rosenbluth (in
http://www.er.doe.gov/production/cyclone/pdf/Marshall's_Paper_V_1B.pdf).
1-K_p is what Dimits compared with in the figure he showed at TTF.

We will soon report on our gyrofluid attempts to reproduce this.  We are
puzzled that for some parameters the gyrofluid model has phi damping to
zero, which is incorrect according to the above picture, while for some
other parameters (or perhaps some initial conditions involving a finite
initial u_parallel) there can be a significant residual undamped phi.  It
might be useful to extend the Hinton-Rosenbluth calculation to include
u_parallel initial conditions.