Ok, I believe I have the solution. It appears that floating point operations for this particular processor is extremely expensive (at least compared to available memory) and any use will increase the binary significantly. For reasons that I really don't understand, if I comment out the *pBuf++ = 0 line, the optimizer is somehow able to optimize out some of the floating point operations in this function and hence the memory is saved.

The bottom line is that, for this processor, I need to avoid floating point at all costs.

Thanks to all who took a look at this.