I tried a benchmark of my own using a function which computes the bearing between two locations specified as lat/long. It produces essentially the same kind of timing as seen before.
In summary, compared to a Duemilanove (=2009) the result of calling the function 10000 times is:
2009 @ 16MHz 12.57secs
T3 @ 48MHz 8.30 secs
T3 @ 96MHz 7.82 secs
The interesting thing I noted was the difference in the size of the code.
The Duemilanove code is 4764 bytes whereas the T3 code is 19984 bytes! I presume that the T3 is hauling in different libraries but it seems to be a bit excessive. OTOH it is 19984 bytes of a 131,072 byte maximum so there's plenty of room to spare 
Pete