the IHS can only reduce the effective removal of heat generated by the die.
I'd have to disagree with such a blanket statement. The IHS is only a detriment if it increases thermal resistance to the heatsink. Yes, I wouldn't be surprised if it currently does in most cases, or at best is merely on-par with a direct interface.
Given that the thickness of copper in the IHS is likely only around 1.5mm thick, its effectiveness in actually 'spreading' heat from the silicon wouldn't be that pronounced - this is part of the reason why the intel stock heatsink just has a circular contact area, making contact in the corners of the IHS doesn't really yield much if any benefit. However, even if an IHS merely doubles the effective heat transfer area it can still be on par with a direct interface if it can provide a better thermal interface to the silicon.
That's really what it all comes down to. Thermal interface materials vary from the 2-3 W/mK that you're likely to find on stock heatsinks to ~9 W/mK for enthusiast materials to the ~20 W/mK of Indigo Extreme... All of which are quite a bit lower than the 149 W/mK of silicon and 390 W/mK of pure copper. If you take the 3 W/mK figure and assume that a 0.05mm joint thickness is possible with heatsink while 0.025mm is possible with IHS and that the IHS lateral heat-spread effectively doubles heat conduction area then you get the following:
Direct contact: 3 W/mK * 0.00016m^2 / 0.00005m = 9.6W/K = 9.9K rise with a 95W load
IHS: 3 W/mK * 0.00016m^2 / 0.000025m = 19.2W/K = 4.95K rise with a 95W load
+ 3 W/mK * 0.00032m^2 / 0.00005 = 19.2W/K = 4.95K rise with a 95W load
= 9.9K rise with 95W load.
Yes, that it's exceedingly simplistic comparison, but it works well to demonstrate why it does make sense to use an IHS and why for non-enthusiast purposes it's not necessarily detrimental. If due to the increasing heat density of future process nodes Intel develops an even better interface between silicon and IHS, then the presence of an IHS could easily become a benefit in all cases.
Edit: Actually, the TIM Intel's using between silicon and IHS is far better than I thought -
http://www.intel.com/technology/itj/2008/v12i1/1-materials/5-solder.htm According to that, the silicon die is effectively soldered to the IHS with Indium, which has a thermal conductivity of 87 W/mK. The date on the article is February 2008 - has anyone actually removed their IHS on an Intel processor and had positive results since then?