Actually you get quite a bit of advantage. The instruction set for 64 bit mode is quite a bit more efficient even when you're not doing 64 bit arithmetic, not least because there are twice as many registers. Code that can be optimised using the SIMD instructions also gets twice as many SIMD registers, which can make a big difference for many graphics, signal processing and mathematical tasks. Most floating point code will be more efficient in 64 bit mode too.
Furthermore the 64 bit address space is useful even with only 1GB of RAM since it allows you to mmap files that are bigger than 4GB. Programs like MongoDB mmap their database files and when running on 32 bit processors MongoDB actually limits processes to a total of 2GB of data in the database, irrespective of how small the working set is. With a 64 bit address space the limit goes from below typical storage sizes to far beyond any plausible storage you'll ever connect to a Pi.