// We have optimized the heap sizes to reduce TLB misses. Change this and you could &%$# up performance (by as much as 1+ ms/frame)

Cache misses are one thing but here it's virtual memory and page tables!?

TLB = Translation Lookaside Buffer