For a long time I had an idea I thought was great - a lock free garbage collector that could be run on its own core to achieve zero latency. I've got 32 cores busy doing nothing and thought I could throw hardware at the problem and get garbage collection for free.

A bunch of people much smarter than me all chimed in on how this was a terrible idea. They argued it would destroy the cache and result in worse performance than a traditional garbage collector. Locality matters, apparently. I still sometimes think of this idea but it doesn't fit very well into the actual hardware.