This site www.lightningsafety.com
contains a lot of info. This appears to be a rather complex issue.
For Brazil, there is this group http://www.cea.inpe.br/webdge/elat/
My instinct tells me to run the grounding lead from this protective device separately down to the electrode(s) in the straightest, shortest route possible. However, even with a very low ground impedance, if the xfmr neutral is bonded to the same electrode system (or even a separate system close by) then the secondary side is still going to rise to a high potential if lightning strikes.
Yes. I did give this some thought, and also looked up how much power there is in a lightning strike and got the figure "up to 100kA". (Don't quote me on this!) This means that even if one has a 1 ohm impedance to earth, the potential of the neutral will jump 1ohm x 100kA = 100 kV (!!!) If the neutral is earthed again between the user and the transformer, the voltage to earth will be reduced as there is a voltage drop in the conductor. But, this also means that you get a DC voltage between the phase and neutral. I suppose that it is this voltage that the surge arrestors are designed to handle.
From the above site:
"Radial horizontal arcing has been measured at least 20 m. from the point where lightning enters the earth."
In theory, you could use the earth to "filter out" the lightning, if the lightning conductor and the secondary side of the transformer are on separate earth electrodes. But they need to be far apart, as can be seen from the above quote.
As we all know, it's preferable and often required to bond all earth conductors to the neutral in LV systems as a fault from phase to unbonded metalwork will not blow the fuse, even if earthed. (I think we can rule out the use of Earth leakage devices in this case.) This is in line with Ricks suggestion.
However, I came to think of a possible option. If the lightning conductor is earthed via an insulated wire, the risk of fault between this wire and phase is considerably limited. This would allow for a earth conductor run to an earth electrode some distance away without bonding to the neutral. (The rest of the transformer setup is bonded to neutral as usual.) The wire needs to be insulated to a voltage at least as high as that which will arise between the two electrode systems. As a first approximation, this will be the impedance to earth times lightning current, which has already been calculated to 100kV DC.