My lesson from examples like this is that there's no point chasing compatibility. I was telling @Pezo - Zoltan Peto this the other day. I'm not familiar with Pypy, but I've seen several OSs chase POSIX-compatibility, starting with OpenBSD. Nobody cares. Inevitably they aren't 100% compatible on the axes people care about, and 99.9% compatibility yields 0.1% adoption. (OpenBSD usage is all by people who live exclusively in OpenBSD.)
And people are right. They're being rational. They're following the design rule that similar things should look similar and different things should look different. If it looks similar but is different under the hood it's in an uncanny valley. It takes a lot to build trust in it.
If you're building something new today, stop trying to support existing software. It's a long, thankless death-march, and will never get you the adoption you want. Take the leap and let it be its own unique thing, without constraining it with the past.
New projects take 10 years (full time) to gain adoption. Stop trying to find shortcuts.