The internal caches of the CPU are orders of magnitude faster than handling a page fault (which means: CPU traps, switch to kernel, VM lookup, issue I/O request

The internal caches of the CPU are orders of magnitude faster than handling a page fault (which means: CPU traps, switch to kernel, VM lookup, issue I/O request, block for completion, switch to user). TFA's context is a larger-than-memory data structure mapped from disk, where heap block size = page size.