Sounds like people are having problems with the Windows version.  I've
been using the Linux version for a few weeks now (the first released
was the final beta).  It's not bad. 

Compiling seems faster than IFC 7.0, error messages are MUCH better,
and it doesn't print those damned subroutine names while compiling.

In the past, I quickly ran into a number of bugs for each
new release of IFC.

This time, only two, and both are minor.  One involved a Compaq/Intel
extension of dubious value (calling a function using
the subroutine CALL name instead of result=name() ), which is
eliminated with -e95.

The other was a crash with ALLOCATABLE structure components.  Well
I figured this wasn't going to really work fully to begin with (and
IFC 7.x didn't support it) so it was just a disappointment.

(yes I submitted all bugs to premier support).

Only major bummer: only the -CB, for bounds checking is supported, at
the moment.  Shape checking and pointer dereferencing checking aren't.
I hear they will be coming back in a future release.

This is the only thing now keeping my with my other compiler, NAG.  In my
experience, NAG has been very accurate on language conformance,
correctness and especially with precise run-time error messages with
all checking turned on. 

And the good part:

testing my own computation, which is a reasonably complex statistical
data analysis which builds significant dynamically allocated tree
structures.

NAG -O3 some gcc options:     28 seconds
ifort -tpp6 -O3 -ipo -stack_temps:  11 seconds

and this is on an Athlon.