V8 allocates a “gigacage” around every wasm memory area on modern 64 bit processors and marks them so the OS will trap on an unexpected access. This is much more efficient than doing a bounds check on every access. These pages are not actually resident.
“Big number is scary” is not a good way to understand performance.
No, but it means there is one less thing we can rely on to identify "process gone whacky".
Plus, admittedly I didn't think about this thoroughly just now, this sounds like a cheap hack for speed. You could still have an OOB access that would land in some valid memory and thus lead to an exploit, right? It's just made very unlikely by making these areas very large and having ASLR. Proper bounds checks wouldn't have that problem.
“Big number is scary” is not a good way to understand performance.