Um... no. It's very different and there is no gap with solder. Fluxless solder applied under the IHS literally binds the IHS to the die directly. Which is why trying to delid a soldered CPU is really risky. If you don't just sand down all the solder or melt it properly, the die can be ripped apart because the solder is bound that strongly to it:
There is no gap... not in Intel's application of a soldered IHS nor in any competent TIM solution for that matter. The literal point of TIM is to eliminate gaps because air is a horrendous conductor of heat.
In contrast, Intel's thermal paste solution has the IHS bound only to the CPU package's PCB via excessive adhesive whereas the die isn't directly connected to the IHS by anything.
Solder TIM solutions are superior because of significantly higher heat conductivity AND a direct connection of the IHS to the die both. Which is also the reason why delidding a CPU with a soldered-on IHS barely produces any improvements whatsoever even if successful (most of the minuscule, potential gains likely to come from running the die naked without the IHS at all rather than a replacement of solder with any commercially-available TIM).
Solder-based TIM solutions actually do their job: Thermal Interface Material. No ands, ifs, or buts; it's nothing at all like the embarrassingly imprecise smothering of excessive adhesive and poor conductivity of Intel's thermal paste solution.
Magic? It's called physics wise guy; basic thermodynamic properties and thermal conductivity to be precise. And no, Ivy Bridge's heat transfer problems are actually solved by solder and Sandy Bridge is not "just easier to cool". Nothing is "just easier to cool", there are reasons. And Ivy Bridge's reason for poor temperatures versus Sandy Bridge is extremely poor thermal conductivity because of a certain gap and significantly less conductive TIM material. That's a fact (one largely discovered by the process and testing of delidding).
Soldered Ivy Bridge-E hexacores run cooler than soldered Sandy Bridge-E hexacores:
The only potentially confounding variable remaining would be thermal density, however the 4960X is actually more dense versus the 3960X (59% of latter's die size) than the 3770K is versus the 2600K (74% of the latter's die size).
Solder vs paste with a gap is the reason for mainstream Ivy Bridge vs Sandy Bridge temperatures before delidding.
No, everyone blames the TIM and the gap because it's the truth. Delidding results, -E platform comparisons, basic thermodynamic properties, and resulting common sense paint a very clear picture.
As for previous chips using TIM, only the lower-end ones and the ancient ones from before modern CPU heat generation levels; not the flagship mainstream offering and its cut-down companion (like our current 6700K/6600K models). Which, incidentally beginning with SB, almost exclusively retain unlocked multipliers and are advertised as overclocking models with some corresponding premiums.
And speaking of those past models:
That Pentium 4 is likely Northwood or Williamette. This was following a time when high-end CPUs were significantly less power hungry and had significantly lower TDP than today therefore they required less effective cooling. It was also before Intel even began soldering IHSs to the CPU at all (Netburst probably being the catalyst).
Intel first started using fluxless solder under their heatspreader with Prescott. 478 Prescott and LGA 775 Pentium 4s were soldered. Also, if you're trying to make a remotely positive point about thermals and good-engineering choices with sub-IHS paste, don't pick Pentium 4s to do it with. Comparing Skylake or Haswell or Ivy Bridge to any Netburst derivative isn't a strong point in favor of paste or Intel's current decisions.
That's a pretty low-end Core 2 and Wolfdale chip, not comparable to the current i5/i7 -K models. The entire E8xxx line based on a bigger and better Wolfdale chip were all soldered. The Conroe equivalents of the modern 6700K/6600K (nicely named E6700/6600) were also soldered. So were the Core 2 Quad Kentsfield equivalents (Q6700/Q6600). And Yorkfield, and Bloomfield, and Lynnfield.