I've used bounds checking and other dynamic analysis tools, and while the architectures are different it's the code that you're checking - in theory you could run bounds checking on any backend and the result would be the same - the code either steps outside its bounds or it does not.
The only complications are addressing more than 4GB of memory space, dealing with pieces of code you can't cross-compile to a 32-bit architecture (64 bit object files for which you have no source, etc), and general 64 bit migration issues (platform specific code such as checking for 0xFFFFFFFF instead of -1)
What other problems are you running into doing bounds checking on your program? Are you unable to compile a 32 bit version?
It's not your ideal solution, certainly, and one should always check the code they're going to run, but in this case you might not have a choice, unless you want to do your own bounds checking (which is a good idea in any case...).