You still can't use the same container image across architectures, even if the important parts are interpreted code. You can package the architecture-independent parts for both x86 and ARM, of course, but as you sort of touch on, that still leaves dependency issues and surprise differences in the platform ("oh look, this is so fast on my mac where there's hardware acceleration for the C library used in this hot path" followed 3 weeks later by, "why is this so slow on our x86 servers?").