This is a benchmarking artifact from not interpolating the variables:
julia> x = SVector{2,Float64}(1,1)
2-element SVector{2, Float64} with indices SOneTo(2):
1.0
1.0
julia> @benchmark SA[$x]
BenchmarkTools.Trial: 10000 samples with 1000 evaluations.
Range (min … max): 1.538 ns … 7.332 ns ┊ GC (min … max): 0.00% … 0.00%
Time (median): 1.551 ns ┊ GC (median): 0.00%
Time (mean ± σ): 1.558 ns ± 0.098 ns ┊ GC (mean ± σ): 0.00% ± 0.00%
▁▃▇█▆▄
▂▄▇██████▃▂▂▂▂▂▁▁▂▂▂▂▂▂▂▂▂▁▁▁▁▁▂▁▁▁▂▂▂▂▂▂▂▁▁▁▁▁▁▁▁▁▂▂▂▂▂▂ ▃
1.54 ns Histogram: frequency by time 1.66 ns <
Memory estimate: 0 bytes, allocs estimate: 0.
julia> @benchmark SVector{1,SVector{2,Float64}}($x)
BenchmarkTools.Trial: 10000 samples with 1000 evaluations.
Range (min … max): 1.751 ns … 4.871 ns ┊ GC (min … max): 0.00% … 0.00%
Time (median): 1.766 ns ┊ GC (median): 0.00%
Time (mean ± σ): 1.783 ns ± 0.128 ns ┊ GC (mean ± σ): 0.00% ± 0.00%
█▇▇▆
▂▂▃▅████▃▂▂▂▂▂▂▁▂▂▂▂▃▄▄▄▂▂▂▂▂▁▁▁▁▁▂▂▂▂▂▂▂▂▂▂▁▁▁▁▂▁▁▁▁▁▁▂▂ ▃
1.75 ns Histogram: frequency by time 1.89 ns <
Memory estimate: 0 bytes, allocs estimate: 0.
Concerning the original question, that is either a bug or some limitation introduced by internals of the representations. I would report it.