*Very* impressive performance gains with release with 2.5! I've noticed that on pure integer calculations (ex: find 10000 decimals of pi using an integer-only algorithm) the speed gain is 7-8x, but definitely impressive! I'll reconsider the option of a JIT for my own interpreters!
I've also noticed that NeonDescent example uses SPRITEOFF statement that is not documented, unless I missed it... Other secret statements ? ;-)