Speed is not a major issue that distinguishes 64-bit applications from 32-bit applications (it was different in early days when 32-bit applications were faster than 16-bit applications, which was a consequence of mathematical operations that required multiple steps of memory access and calculations on 16-bit computers and even more on 8-bit systems).
The main advantage of a 64-bit program is that it can handle much larger chunks of memory. This is, however, for most applications not important, as memory requirements are not so high that they can't be fulfilled by a 32-bit processor. Therefore, creating a 32-bit application is for most situations the better choice (also with respect to backward-compatibility, as mentioned by xenblaise).
Only for certain requirements, e.g. supercomputing, large-scale computational statistics or data-warehousing, a 64-bit program may be favourable.