OK. I also saw that point in the Photomatix manual some time ago. However:
1) Photomatix's internal raw converter is dcraw, which is certainly not as capable at corrections as ACR, but is a pretty widely used and respected converter. In fact, there are some threads on this forum discussing the use of (and the reasons for use of) dcraw.
2) It may well be possible that the Nikon raw converter is better than non-Nikon converters due to proprietary knowledge by Nikon about its sensors, however, unless you use a specific camera profile, why would you think that a completely zero-ed out ACR would be better than dcraw on NEFs? (I'm not saying it isn't, I would just like to know if you know of some specific reason that it actually is better.)
3) I doubt if you are avoiding the linear/gamma conversion round trip by using a linear curve. The gamma of whatever color space you use in the tiff will be applied as the file is saved, no matter what corrections you asked for. In any case, how would Photomatix know not to apply the gamma->linear conversion on reading the tiff?
4) I did a VERY small experiment something like a year ago. I tried several of my HDRs both ways. I got 1 I liked a little better using dcraw, and something like 3 or 4 where the difference was insignificant.