Wiring code into RAM prevents it from being swapped to storage, so this might be a nice design tradeoff to wire critical components and frameworks to RAM so that the chance of a drive that fails to write data correctly will bring the system down.
Trading speed for stability would be a nice thing for safe mode to enhance it's ability to run in edge cases (drive totally full, drive failing, major software corruption). Even if this is not intentional, the safe boot does throw away (and does not recreate) the dynamic loader shared cache and that also might be a design reason to just pin each library needed into RAM. You would want to run
sysdiagnose before, during and after a safe boot and might be able to see in memory allocation statistics if this makes up for your measured difference in RAM allocation.
I can't speak for the designers, but I would consider making this tradeoff if I were in their shoes and wanted to make Safe Mode as safe to run as possible.