A game engine is more than its ability to push the GPU, and to be blunt this belief is why the video game industry is so stagnant but for a few innovators.
The Source engine in its time was the gold standard for lip synching and emoting. When the unchallenged status quo at the time was flapping mouths Valve implemented full real-time phoneme lip synching as well as dynamic facial expressions - as well as the framework to do so under contextual cues from the game world.
HL2 was heralded in its time for being able to tell stories at a deeper level because of this technology, and there was a great deal of hubbub around it, even in the mainstream press.
Hell, even if we move away from artsy-fartsy things like emoting in gaming, Valve for a long time was the gold standard for their network engine. Following the basic network model as the original Quake engines Valve was able to achieve incredible reliability in client-side prediction and latency management - something most games in HL2's time were unable to do (remember every Battlefield before Battlefield 3?). There's a reason why major competitive multiplayer games grew out of that engine.
So sure, Valve's engine can't render 1000 animated point light sources all at once, nor do they claim to do pixel-perfect real-time shadowing for an entire forest's worth of trees, but to say that Valve's engines aren't a huge step forward in gaming each time they are released is shortsighted.