Question: What makes this call with evalhf slower than the call without evalhf

I have expected the opposite. Is exp already optimised that hardware floats do not make sense or does the conversion of the argument to hardware floast eats up all the benefit of using hardware floats?

restart;
CodeTools:-Usage( for i from 1 to 100 by 0.1 do exp(i) end do):
CodeTools:-Usage( for i from 1 to 100 by 0.1 do (evalhf@exp)(i) end do):
memory used=1.54MiB, alloc change=0 bytes, cpu time=16.00ms, real time=15.00ms, gc time=0ns
memory used=5.88MiB, alloc change=32.00MiB, cpu time=31.00ms, real time=33.00ms, gc time=0ns
Please Wait...