Looks interesting, but it's difficult to understand why it is faster (i.e. the design decisions you made).

From poking around, I can see that you have specialized each collection for the primitive types, to avoid boxing/unboxing. I'm not sure if there's more to it than this though!

I wasn't able to quickly locate the actual storage code though - do you store values as a primitive array or do you use a ByteBuffer? It would also be useful to understand how this compares to e.g. Trove.

Just some friendly feedback - project looks awesome; thanks for open-sourcing it.