I have a notebook with a Skylake generation equivalent part (i5-6267U) sporting an Iris 550 iGPU with 64MB of eDRAM.
At 35 Watts effective at the wall plug it gives the exact same performance as an AMD A10-7850K APU, which has up to 120 Watts at the wall on every graphics and CPU benchmark I've come across: It's really uncanny just how similar performance these two rather distinct architectures have, almost as if Intel was using it as benchmark.
That has seriously impressed me of what the bigger Intel iGPUs can do and made me look for similar ones in newer devices.
Alas, as you found, these simply don't exist.
AFAIK these CPUs are basically Apple custom chips, so you can get them easily as an Apple notebook, after selling a kidney.
Intel then puts some of them (surplus?) into NUCs and at prices that are basically insane, when you consider the relative complexity and power of these chips. Even a normal GT2 chip already uses more die area for the GPU than the CPU cores. But at GT3 or GT3e (65Watt chips they did for Sky and Kaby-Lake), that area doubles and triples. And then they even put 64 or 128MB of eDRAM on the same die carrier to offset the lack of bandwidth in normal DRAM for GPUs.
Yet they sell them in a NUC at a price practically identical to an ordinary GT2 as if the GT3 GPU and eDRAM overhead was zero extra cost. And actually zero (ok $5) is what Intel has been charging even for normal GT2 iGPUs in the few cases they sold CPUs with deactivated iGPUs (Xeons mostly). At 50% surface area on a chip that goes easily for $500 (and with no proportional differentiation for dual core dies), that doesn't make any economical sense from a fab perspective where the size of the chip determines production cost.
All that to my knowledge has never been publically discussed and received far too little attention, because the only explanation that I can come up with is anti-competitive behaviour against AMD and Nvidia.
It's also exactly how the original Ryzen striked back at Intel, converting the iGPU die area into 4 extra CPU cores instead.
Again, the only exception from Apple and NUC I ever found was this Medion Notebook that I bought at quite a competitive price (€600 with VAT) in 2016. It's a standard Pegatron design unchanged for 28 Watt that also supports 15 Watt CPUs and even discrete graphics on the same motherboard and chassis. And the CPU really is only 15Watt TDP like every other U-variant and the gap to 28 Watts is only available to the iGPU: It doesn't clock higher or longer than an equivalent 15 Watt Skylake cousin. It's not a 28 Watt CPU only a 28 Watt SoC. There are some benchmark benefits for the CPU, as the eDRAM is in fact not exclusive to the GPU but an L4 cache usable by both, but tiny.
How that SoC slipped through a barrier that Intel and Apple have held high until the latest Cannon-Lake announcements (where the need for eDRAM was compensated by faster DDR4), I don't know. But like you I've always wanted a GT3 laptop at a GT2 price. And I guess the ones from Apple show that the hardware core is attractive enough in terms of electrical and compute/graphics power balance.
It doesn't really give you a gaming machine, though. And the production cost of these chips most likely weren't competitive against entry level dGPUs.
Too bad nobody makes a notebook chassis that you can simply slip a NUC motherboard into. Not everybody needs an ultrabook!