Elements like Mg, Si, Al, Ca, Ti, Fe, Ni are concentrated in interstellar dust grains. Since dust grains can move through the gas, transport in dust grains could produce local variations in elemental abundances.
Starlight is typically anisotropic, as can be confirmed by viewing the night sky on a clear night. The anisotropy is a function of wavelength, with larger anisotropies at shorter wavelengths because UV is (1) more strongly attenuated by dust grains and (2) originates in a smaller number of short-lived stars that are clustered. At the location of the Sun, Weingartner & Draine (2001b) found starlight anistropies ranging from 3% at 5500Å to 21% at 1565Å.
The dust grains are charged, and fairly well-coupled to the magnetic
field lines. Let be the angle between the starlight anisotropy direction
and the magnetic field.
The drift tends to be approximately parallel to the magnetic field
direction, with a magnitude
vd 
vd0 cos
B; for a
starlight anisotropy of 10%,
vd0
0.5 km s-1 in the "warm neutral medium", and
0.03 km s-1 in the "cold neutral medium"
(Weingartner & Draine
2001b).
If the radiation anisotropy is stable for ~ 107 yr, a dust grain
in the cold neutral medium would be driven
0.3 cos B pc
from the gas element in which it was originally located.
If the magnetic field is nonuniform, spatial variations in the drift velocity can then lead to variations in the dust/gas ratio. Indeed, if there are "valleys" in the magnetic field (with respect to the direction of radiation anisotropy, the dust grains would tend to be concentrated there. Radiation pressure acting on the concentrated dust grains would in fact cause such perturbations in the magnetic field to be unstable.
Dust impact detectors on the Ulysses and Galileo spacecraft measure
the flux and mass distribution of
interstellar grains entering the solar system, finding
a much higher flux of very large (a
0.5µm dust grains than would be expected for the average
interstellar grain size distribution
(Frisch et al. 1999).
The inferred local dust size distribution is very difficult to
reconcile with the average interstellar grain population
(Weingartner & Draine
2001a),
but we must keep in mind that the dust grains entering the solar system
over a time scale of a few years are sampling a
tiny region of the interstellar medium of order a few AU in size.
It is possible that the solar system just happens to be passing
through a region in the local interstellar cloud which has been
enhanced in the abundance of large grains due to radiation-driven
grain drift.
Various processes could alter the gas-to-dust ratio in the material forming a star:
Motion of dust through gas can result from gravitational sedimentation of dust grains in star-forming clouds (Flannery & Krook 1978). This could lead to enhanced abundances in stars of those elements which are depleted into dust in star-forming clouds.
Star formation is accompanied by ambipolar diffusion of magnetic field out of the contracting gas cloud. Charged dust grains, while not perfectly coupled to the magnetic field, will tend to drift outward, resulting in reduction of the abundances of the depleted elements in the star-forming core (Ciolek & Mouschovias 1996).
Gravitational sedimentation in an accretion disk can concentrate dust at the midplane. Gammie (1996) has suggested "layered accretion", in which viscous stresses are effective along the surface layers of the disk, but the midplane is a quiescent "dead zone" as far as the magnetorotational instability is concerned. This could suppress stellar abundances of depleted elements.
In thick disks, small bodies orbit more rapidly than the gas, with gas drag leading to radial infall of these bodies. This could enhance stellar abundances of depleted species.
Before accretion terminates, a massive star may attain a pre-main-sequence luminosity high enough for radiation pressure to drive a drift of dust grains away from it. This would suppress stellar abundances of depleted elements.
Given these competing mechanisms, the net effect could be of either sign, but, as pointed out by Snow (2000), it would not be suprising if stars had abundances which differed from the overall abundances of the interstellar medium out of which they formed, and it would also not be surprising if the resulting abundances depended on stellar mass. In this connection it is interesting to note that the study by Sofia & Meyer (2001) finds that abundances of Mg and Si in B stars appear to be significantly below the abundances of these elements in young F and G disk stars, or in the Sun; they conclude that the abundances in B stars do not provide a good representation of interstellar abundances.