Hey that is a great graph. But if you stack the graph up next to the LL test times by the same machines you will find that AMD64 and Pentium-M are indeed more suitable. Particularly if we take into account the new factoring limits which mean that less time is spent on the bits above 64 which reduce the advantage of the AMD64 and P4.

Still, as the graph clearly shows, the AMD64 is teh 1ner.

Well, if we are talking about LMH ranges this will typically be, also in future, a factoring depth between 61 and 63 bits, ie. the following TF before the LL-test will take most of the time (>90%) for factoring from 64 to 67 bits or higher.
Nevertheless, everybody can do with his computer what he wants to do.
Maybe I'll add the graph of the LL timings for the same procs later, just as an additional for help for decision.