Dear Torsten,
your numbers are definitely helpful in this discussion.
In our experience it has proven very difficult to prepare hydrogen from water
using zink (BDH, Hayes) without fractionation. Numbers of SLAP in excess of -420
are considered good, provided they are reproducible. In addition the zink
reaction is very sensitive towards contaminants.
We recently came up with an automated equilibration system. Here, due to the
strong fractionation at 20C, the absolute mass 3 contribution is only 78 ppm
thus rendering these measurements more difficult from the mass spec view. The
H3+ correction (around 10 ppm/nA, giving a 50 ppm contribution to the mass 3
channel for a typical 5 nA signal) is the same, but the relative error is
increased 4fold. For these measurements we got an average SMOW/SLAP difference
of -426.3 permil for a wide variety of working standards (-800 to + 92 permil vs
SMOW). The expected value is close to -427 permil due to the mixing effect of
the Standard water plus H2 gas.
Since the H2 gas equilibrated with SMOW is around -750 permil, most of the
reference gases were outside the scale in the positive direction. Systematic
errors of the MS (252) would have been prominent.
Another set of measurements was made following the procedure of the people in
Leipzig who use a chromium alloy in their reduction reactor. Again, the values
for the SM/SL difference were close to -428 (-428.6 in this case, s.d. 0.8
permil on my Delta S) without any further data treatment.
Therefore I would like to ask the question: How much of your results do you
think are attributable to sample preparation rather than an MS effect? 
The MS effects I can think of are the following:
1. H3+ production from proton donators other than H2+ (CH4+, H2O+). Cure: check
for these impurities, treat all samples equal. This effect would require
scaling.
2. Poor H3+ contribution stability. The determination of the H3+ factor is
limited in precision. Provided that no other cause for variability of data is
present, the maximum allowance for a 1 permil reproduciblility e.g. on
equilibrated hydrogen is 1.6 permil of the 10 ppm/nA. This is 10 +- 0.016
ppm/nA. Therefore the determination of the H3 factor should be made only once in
a series of data. Minor scaling is necessary.
3. Poor secondary electron suppression. Any deviation of the detection
efficiency from exactly 1 charge per incoming ion will lead to a wrong value.
The direction of this effect is not predictable, depending on whether secondary
electrons from external wall collisions can enter the cup or whether secondary
electrons from inside the cup can leave it. A high quality electron barrier at
the cup entry to my knowledge is the only solution to this effect. It is the aim
of our MS design to make this effect neglegible. For mass specs with permanent
magnets as suppressors, I would check this effect with gases of known isotopic
compositions very carefully.
Regards	Willi Brand