There was a presentation by Andrew Kelley, creator and lead developer of the Zig programming language, about size optimisations he's applying within the Zig compiler. He goes further than just reordering fields in order to achieve even greater gains, but it's all about shrinking the size of values and reducing padding.
At 38:15 he gives some numbers after applying these techniques to the reduce the size of tokens and AST nodes:
Spoilers
- 15% fewer cache-misses
- 28% fewer total instructions executed
- 26% fewer total CPU cycles
- 22% faster wall clock time
And then at 42:30 he gives some numbers after reducing the size of Zig IR instruction values:
Spoilers
- 39% reduction in Wall Clock Time
- 40% reduction in Peak Memory Usage
- 53% reduction in Cache Misses
- 23% reduction in Instructions
- 41% reduction in CPU Cycles
(these are in addition to the previous gains)
Your results will of course depend on the specifics of your application.