Besides your claim of AV1 encoding being "extremely resource demanding" being out-of-date (that hasn't been the case for years), this is kind of like asking for proof that the sky is really blue. Video coding is a niche enough field that academics generally don't write papers on "is the sky blue" type questions, and amateur video encoder tests (when they do exist) are by-and-large terrible because the people conducting them often have no idea what they're doing. Reliable data is so hard to find that the most common and useful answer to a question like this is "do your own testing for your specific use case".
If you do look at research on hardware encoders though, you may see a bunch of "comparable to software encoding" claims. However, these are for specifically designed implementations, and the reality of practical hardware encoders is that they rarely have the luxury of only focusing on a single video coding format. They need to be able to produce H.264 and HEVC, and possibly others as well. Combine this with a strict silicon area budget of consumer hardware and you get multi-format hardware encoders that need to have as many shared parts as possible. With all these limitations, it's easier understand why most hardware encoders only target the lower end of the efficiency curve, and that hardware encoders with the "same settings" as software encoders don't really exist. For example, no AV1 hardware encoder I currently know of has implemented film grain synthesis, so hardware encoders may completely skip some encoding tools, which will reduce efficiency.
This isn't to say that there are no hardware encoders with higher-than-average efficiency. When most people say "hardware encoder", what they really mean is "hardware encoder I could get my hands on". Dedicated hardware encoder cards aimed for commercial/datacenter users do exist, for example this AMD card which claims x265 slow efficiency for its AV1 encoder (roughly comparable to SVT-AV1's current preset 8, so still not comparable to what software encoders can produce). Companies that deal with video at large enough scale may also design their own hardware encoders for private use, which is something Google has done.
If you really are that skeptical about hardware encoder efficiency and you're interested enough, I suggest you prove this to yourself. Consumer-grade AV1 hardware encoders are easy enough to obtain. There are some very affordable Intel Arc cards, but Nvidia & AMD cards with AV1 encode are currently more expensive. Or you could try literally any hardware encoder for any format, since the principle is exactly the same. There's nothing special about AV1 hardware encoders that make them fundamentally different in this context than, say, H.264 or HEVC hardware encoders. However, some recent encoders for older formats may (or should, even) be better in their comparative efficiency than these first-generation AV1 hardware encoders, since the former are on their Nth hardware iteration and have therefore had more development time.
As a bit of a meta-aside, this question seems to come up every time a new video compression format is introduced and the initial software implementations are slow compared to encoders for current-generation formats. Hardware encoders get hyped up by some as this amazing future technology that will save us from single-digit encoding speeds, but the end result is always the same. Hardware encoders don't reach anywhere near the full potential of video compression formats, and software development, optimization, and more powerful CPUs are what actually bring us the most efficient encoders. So, for the information of OP and anyone reading this in the future: hype and misleading/false claims are not uncommon when it comes to video encoders. In fact, they're pretty much the rule rather than an exception.
