This presentation is strange, that's for sure. Hard to believe the equivalent measures are being compared.

As the comparisons between P-III and P4 double libraries confirm, you would expect more clocks to be required when x87 extended precision is used to obtain a fully accurate result. Also confirmed is an increased number of clocks for the latest style of P4, compared with the earlier one.

I assume the distinction between the 2 columns is according to whether the 32- or 64-bit library is in use, so you could take your choice on Pentium-D as well. I suppose you could run all versions of the library on your Pentium-D with 64-bit OS, and check the tables yourself.

VML part of MKL 7.2 has several functions, which are not optimized for EM64T.
The reason is that initially Microsoft announced to drop x87 support in the OS.
So we were forced to quickly get rid of x87-based functions by temporarily replacing
them with rather slow portable implementations.
Improvements for some of these functions will be in MKL 8.0
(e.g. for Tan, Atan2, Asinh, Acosh for certain floating-point format and accuracy flavor)
while for other remaining functions it will be done in subsequent releases.

As to your second question about VML performance data for Pentium D processor,
we do not publish this data on the web yet.
Im not sure that existing performance columns will help you getting absolutely
adequate performance information on Pentium D processor.

Thanks,
Andrey