It has been clear from the very beginning that Intel's efficiency core is all about space efficiency and not power efficiency. In the silicon space where they would previously fit 1 Sunnycove core with hyperthreading, they can fit 4 atom cores without hyperthreading. Hence a quad-core TigerLake becomes a 2 cores 4 threads TigerLake plus 8 core atom, known as AlderLake 10/12 a.k.a. AlderLake-U.
They are doing it for core stacking. Atom cores are good at integer workloads, matching big cores at 2:1 ratio (2 atom cores = 1 Skylake w/ HT). So compared to the quad-core TigerLake, they now perform like a hexa-core with AlderLake-U on the same silicon. A hexa TigerLake now performs likes an octa with AlderLake-P.
But atom cores are purpose built cores that only prioritize integer performance. It neglects all the other x86 extensions that are nice to have, like AVX-256, AVX-512, SHA, etc. The extra stuff take more silicon space and have more leakage, which is why big cores are so complex and inefficient. If you benchmark AlderLake on AVX-2, the performance advantage falls apart.