Due to multilevel caches I doubt this would be so applicable to modern systems - the index-set representation takes several times more memory and naturally requires lots of random accesses, which are bad for caches. In some ways, it's no longer a time-space tradeoff: smaller is faster.

Also, initialising the space doesn't take so long; REP STOS on x86 is fast since it can write an entire cache line at once: http://ptspts.blogspot.ca/2014/10/on-speed-of-memset.html