Heh, yea, that tends to happen. I wrote a precaching system for my game to avoid what I thought was a long series of loops that works in a trial-and-error way, just to find when testing that all the loops took 0.1 seconds to complete!

Oh, I prefer

exception

when others

then null;

as the root of all evil. But meh, I code in PL/SQL.

I have had this issue before... but optimizing for debug mode is nice if it means the game is playable in debug mode (great for tracking down reproducible crashes). 

It would still be interesting to know the root cause of why the optimized code was so many orders of magnitude faster than debug.  Especially if it has to do with DirectX, since I plan on diving into CUDA before too long.  Knowing that bottleneck might give clues to other performance problems, particularly if it involves some sort of unnecessary and complex handshake between the CPU and GPU.