Question: are there measurable qualities of untrusted WASM code (like size, structure, etc.) which could either greatly reduce or entirely eliminate the potential of an explosion in cross-compilation time?

One freebie is the ability to run the untrusted code in interpreted mode many times before attempting any cross-compiling.

Are there domain specific languages, or language subsets (subsets of WASM?) which might offer this fast and safe cross-compilation automatically?

Background: I'm interested in a system where machines of potentially arbitrary architecture take in and run untrusted WASM code. Putting sandboxing aside, my primary concern is 'compiler bombs' where a WASM program heavily burdens the AOT or JIT translation process in an unpredictable way (predictable and universal (across architectures) explosions in run-time are acceptable).

Note that speed is also a primary goal, despite running untrusted code.